Upload folder using huggingface_hub

.gitattributes

@@ -34,3 +34,18 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 diffsynth/tokenizer_configs/hunyuan_video/tokenizer_2/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+diffsynth/models/__pycache__/sd3_text_encoder.cpython-311.pyc filter=lfs diff=lfs merge=lfs -text
+diffsynth/models/__pycache__/sd_unet.cpython-311.pyc filter=lfs diff=lfs merge=lfs -text
+diffsynth/models/__pycache__/sdxl_unet.cpython-311.pyc filter=lfs diff=lfs merge=lfs -text
+diffsynth/models/__pycache__/svd_unet.cpython-311.pyc filter=lfs diff=lfs merge=lfs -text
+diffsynth/tokenizer_configs/kolors/tokenizer/vocab.txt filter=lfs diff=lfs merge=lfs -text
+models/Wan-AI/Wan2.1-T2V-1.3B/assets/comp_effic.png filter=lfs diff=lfs merge=lfs -text
+models/Wan-AI/Wan2.1-T2V-1.3B/assets/data_for_diff_stage.jpg filter=lfs diff=lfs merge=lfs -text
+models/Wan-AI/Wan2.1-T2V-1.3B/assets/i2v_res.png filter=lfs diff=lfs merge=lfs -text
+models/Wan-AI/Wan2.1-T2V-1.3B/assets/t2v_res.jpg filter=lfs diff=lfs merge=lfs -text
+models/Wan-AI/Wan2.1-T2V-1.3B/assets/vben_1.3b_vs_sota.png filter=lfs diff=lfs merge=lfs -text
+models/Wan-AI/Wan2.1-T2V-1.3B/assets/vben_vs_sota.png filter=lfs diff=lfs merge=lfs -text
+models/Wan-AI/Wan2.1-T2V-1.3B/assets/video_dit_arch.jpg filter=lfs diff=lfs merge=lfs -text
+models/Wan-AI/Wan2.1-T2V-1.3B/assets/video_vae_res.jpg filter=lfs diff=lfs merge=lfs -text
+models/Wan-AI/Wan2.1-T2V-1.3B/examples/i2v_input.JPG filter=lfs diff=lfs merge=lfs -text
+models/Wan-AI/Wan2.1-T2V-1.3B/google/umt5-xxl/tokenizer.json filter=lfs diff=lfs merge=lfs -text

.github/workflows/publish.yaml

@@ -0,0 +1,29 @@
+name: release
+
+on:
+  push:
+    tags:
+      - 'v**'
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}-publish
+  cancel-in-progress: true
+
+jobs:
+  build-n-publish:
+    runs-on: ubuntu-20.04
+    #if: startsWith(github.event.ref, 'refs/tags')
+    steps:
+      - uses: actions/checkout@v2
+      - name: Set up Python 3.10
+        uses: actions/setup-python@v2
+        with:
+          python-version: '3.10'
+      - name: Install wheel
+        run: pip install wheel && pip install -r requirements.txt
+      - name: Build DiffSynth
+        run: python setup.py sdist bdist_wheel
+      - name: Publish package to PyPI
+        run: |
+          pip install twine
+          twine upload dist/* --skip-existing -u __token__ -p ${{ secrets.PYPI_API_TOKEN }}

LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [2023] [Zhongjie Duan]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

README.md

@@ -0,0 +1,244 @@
+# DiffSynth Studio
+[![PyPI](https://img.shields.io/pypi/v/DiffSynth)](https://pypi.org/project/DiffSynth/)
+[![license](https://img.shields.io/github/license/modelscope/DiffSynth-Studio.svg)](https://github.com/modelscope/DiffSynth-Studio/blob/master/LICENSE)
+[![open issues](https://isitmaintained.com/badge/open/modelscope/DiffSynth-Studio.svg)](https://github.com/modelscope/DiffSynth-Studio/issues)
+[![GitHub pull-requests](https://img.shields.io/github/issues-pr/modelscope/DiffSynth-Studio.svg)](https://GitHub.com/modelscope/DiffSynth-Studio/pull/)
+[![GitHub latest commit](https://badgen.net/github/last-commit/modelscope/DiffSynth-Studio)](https://GitHub.com/modelscope/DiffSynth-Studio/commit/)
+
+<p align="center">
+<a href="https://trendshift.io/repositories/10946" target="_blank"><img src="https://trendshift.io/api/badge/repositories/10946" alt="modelscope%2FDiffSynth-Studio | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a>
+</p>
+
+Document: https://diffsynth-studio.readthedocs.io/zh-cn/latest/index.html
+
+## Introduction
+
+DiffSynth Studio is a Diffusion engine. We have restructured architectures including Text Encoder, UNet, VAE, among others, maintaining compatibility with models from the open-source community while enhancing computational performance. We provide many interesting features. Enjoy the magic of Diffusion models!
+
+Until now, DiffSynth Studio has supported the following models:
+
+* [Wan-Video](https://github.com/Wan-Video/Wan2.1)
+* [StepVideo](https://github.com/stepfun-ai/Step-Video-T2V)
+* [HunyuanVideo](https://github.com/Tencent/HunyuanVideo)
+* [CogVideoX](https://huggingface.co/THUDM/CogVideoX-5b)
+* [FLUX](https://huggingface.co/black-forest-labs/FLUX.1-dev)
+* [ExVideo](https://huggingface.co/ECNU-CILab/ExVideo-SVD-128f-v1)
+* [Kolors](https://huggingface.co/Kwai-Kolors/Kolors)
+* [Stable Diffusion 3](https://huggingface.co/stabilityai/stable-diffusion-3-medium)
+* [Stable Video Diffusion](https://huggingface.co/stabilityai/stable-video-diffusion-img2vid-xt)
+* [Hunyuan-DiT](https://github.com/Tencent/HunyuanDiT)
+* [RIFE](https://github.com/hzwer/ECCV2022-RIFE)
+* [ESRGAN](https://github.com/xinntao/ESRGAN)
+* [Ip-Adapter](https://github.com/tencent-ailab/IP-Adapter)
+* [AnimateDiff](https://github.com/guoyww/animatediff/)
+* [ControlNet](https://github.com/lllyasviel/ControlNet)
+* [Stable Diffusion XL](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
+* [Stable Diffusion](https://huggingface.co/runwayml/stable-diffusion-v1-5)
+
+## News
+
+- **February 25, 2025** We support Wan-Video, a collection of SOTA video synthesis models open-sourced by Alibaba. See [./examples/wanvideo/](./examples/wanvideo/).
+
+- **February 17, 2025** We support [StepVideo](https://modelscope.cn/models/stepfun-ai/stepvideo-t2v/summary)! State-of-the-art video synthesis model! See [./examples/stepvideo](./examples/stepvideo/).
+
+- **December 31, 2024** We propose EliGen, a novel framework for precise entity-level controlled text-to-image generation, complemented by an inpainting fusion pipeline to extend its capabilities to image inpainting tasks. EliGen seamlessly integrates with existing community models, such as IP-Adapter and In-Context LoRA, enhancing its versatility. For more details, see [./examples/EntityControl](./examples/EntityControl/).
+  - Paper: [EliGen: Entity-Level Controlled Image Generation with Regional Attention](https://arxiv.org/abs/2501.01097)
+  - Model: [ModelScope](https://www.modelscope.cn/models/DiffSynth-Studio/Eligen), [HuggingFace](https://huggingface.co/modelscope/EliGen)
+  - Online Demo: [ModelScope EliGen Studio](https://www.modelscope.cn/studios/DiffSynth-Studio/EliGen)
+  - Training Dataset: [EliGen Train Set](https://www.modelscope.cn/datasets/DiffSynth-Studio/EliGenTrainSet)
+
+- **December 19, 2024** We implement advanced VRAM management for HunyuanVideo, making it possible to generate videos at a resolution of 129x720x1280 using 24GB of VRAM, or at 129x512x384 resolution with just 6GB of VRAM. Please refer to [./examples/HunyuanVideo/](./examples/HunyuanVideo/) for more details.
+
+- **December 18, 2024** We propose ArtAug, an approach designed to improve text-to-image synthesis models through synthesis-understanding interactions. We have trained an ArtAug enhancement module for FLUX.1-dev in the format of LoRA. This model integrates the aesthetic understanding of Qwen2-VL-72B into FLUX.1-dev, leading to an improvement in the quality of generated images.
+  - Paper: https://arxiv.org/abs/2412.12888
+  - Examples: https://github.com/modelscope/DiffSynth-Studio/tree/main/examples/ArtAug
+  - Model: [ModelScope](https://www.modelscope.cn/models/DiffSynth-Studio/ArtAug-lora-FLUX.1dev-v1), [HuggingFace](https://huggingface.co/ECNU-CILab/ArtAug-lora-FLUX.1dev-v1)
+  - Demo: [ModelScope](https://modelscope.cn/aigc/imageGeneration?tab=advanced&versionId=7228&modelType=LoRA&sdVersion=FLUX_1&modelUrl=modelscope%3A%2F%2FDiffSynth-Studio%2FArtAug-lora-FLUX.1dev-v1%3Frevision%3Dv1.0), HuggingFace (Coming soon)
+
+- **October 25, 2024** We provide extensive FLUX ControlNet support. This project supports many different ControlNet models that can be freely combined, even if their structures differ. Additionally, ControlNet models are compatible with high-resolution refinement and partition control techniques, enabling very powerful controllable image generation. See [`./examples/ControlNet/`](./examples/ControlNet/).
+
+- **October 8, 2024.** We release the extended LoRA based on CogVideoX-5B and ExVideo. You can download this model from [ModelScope](https://modelscope.cn/models/ECNU-CILab/ExVideo-CogVideoX-LoRA-129f-v1) or [HuggingFace](https://huggingface.co/ECNU-CILab/ExVideo-CogVideoX-LoRA-129f-v1).
+
+- **August 22, 2024.** CogVideoX-5B is supported in this project. See [here](/examples/video_synthesis/). We provide several interesting features for this text-to-video model, including
+  - Text to video
+  - Video editing
+  - Self-upscaling
+  - Video interpolation
+
+- **August 22, 2024.** We have implemented an interesting painter that supports all text-to-image models. Now you can create stunning images using the painter, with assistance from AI!
+  - Use it in our [WebUI](#usage-in-webui).
+
+- **August 21, 2024.** FLUX is supported in DiffSynth-Studio.
+  - Enable CFG and highres-fix to improve visual quality. See [here](/examples/image_synthesis/README.md)
+  - LoRA, ControlNet, and additional models will be available soon.
+
+- **June 21, 2024.** 🔥🔥🔥 We propose ExVideo, a post-tuning technique aimed at enhancing the capability of video generation models. We have extended Stable Video Diffusion to achieve the generation of long videos up to 128 frames.
+  - [Project Page](https://ecnu-cilab.github.io/ExVideoProjectPage/)
+  - Source code is released in this repo. See [`examples/ExVideo`](./examples/ExVideo/).
+  - Models are released on [HuggingFace](https://huggingface.co/ECNU-CILab/ExVideo-SVD-128f-v1) and [ModelScope](https://modelscope.cn/models/ECNU-CILab/ExVideo-SVD-128f-v1).
+  - Technical report is released on [arXiv](https://arxiv.org/abs/2406.14130).
+  - You can try ExVideo in this [Demo](https://huggingface.co/spaces/modelscope/ExVideo-SVD-128f-v1)!
+
+- **June 13, 2024.** DiffSynth Studio is transferred to ModelScope. The developers have transitioned from "I" to "we". Of course, I will still participate in development and maintenance.
+
+- **Jan 29, 2024.** We propose Diffutoon, a fantastic solution for toon shading.
+  - [Project Page](https://ecnu-cilab.github.io/DiffutoonProjectPage/)
+  - The source codes are released in this project.
+  - The technical report (IJCAI 2024) is released on [arXiv](https://arxiv.org/abs/2401.16224).
+
+- **Dec 8, 2023.** We decide to develop a new Project, aiming to release the potential of diffusion models, especially in video synthesis. The development of this project is started.
+
+- **Nov 15, 2023.** We propose FastBlend, a powerful video deflickering algorithm.
+  - The sd-webui extension is released on [GitHub](https://github.com/Artiprocher/sd-webui-fastblend).
+  - Demo videos are shown on Bilibili, including three tasks.
+    - [Video deflickering](https://www.bilibili.com/video/BV1d94y1W7PE)
+    - [Video interpolation](https://www.bilibili.com/video/BV1Lw411m71p)
+    - [Image-driven video rendering](https://www.bilibili.com/video/BV1RB4y1Z7LF)
+  - The technical report is released on [arXiv](https://arxiv.org/abs/2311.09265).
+  - An unofficial ComfyUI extension developed by other users is released on [GitHub](https://github.com/AInseven/ComfyUI-fastblend).
+
+- **Oct 1, 2023.** We release an early version of this project, namely FastSDXL. A try for building a diffusion engine.
+  - The source codes are released on [GitHub](https://github.com/Artiprocher/FastSDXL).
+  - FastSDXL includes a trainable OLSS scheduler for efficiency improvement.
+    - The original repo of OLSS is [here](https://github.com/alibaba/EasyNLP/tree/master/diffusion/olss_scheduler).
+    - The technical report (CIKM 2023) is released on [arXiv](https://arxiv.org/abs/2305.14677).
+    - A demo video is shown on [Bilibili](https://www.bilibili.com/video/BV1w8411y7uj).
+    - Since OLSS requires additional training, we don't implement it in this project.
+
+- **Aug 29, 2023.** We propose DiffSynth, a video synthesis framework.
+  - [Project Page](https://ecnu-cilab.github.io/DiffSynth.github.io/).
+  - The source codes are released in [EasyNLP](https://github.com/alibaba/EasyNLP/tree/master/diffusion/DiffSynth).
+  - The technical report (ECML PKDD 2024) is released on [arXiv](https://arxiv.org/abs/2308.03463).
+
+
+## Installation
+
+Install from source code (recommended):
+
+```
+git clone https://github.com/modelscope/DiffSynth-Studio.git
+cd DiffSynth-Studio
+pip install -e .
+```
+
+Or install from pypi (There is a delay in the update. If you want to experience the latest features, please do not use this installation method.):
+
+```
+pip install diffsynth
+```
+
+If you encounter issues during installation, it may be caused by the packages we depend on. Please refer to the documentation of the package that caused the problem.
+
+* [torch](https://pytorch.org/get-started/locally/)
+* [sentencepiece](https://github.com/google/sentencepiece)
+* [cmake](https://cmake.org)
+* [cupy](https://docs.cupy.dev/en/stable/install.html)
+
+## Usage (in Python code)
+
+The Python examples are in [`examples`](./examples/). We provide an overview here.
+
+### Download Models
+
+Download the pre-set models. Model IDs can be found in [config file](/diffsynth/configs/model_config.py).
+
+```python
+from diffsynth import download_models
+
+download_models(["FLUX.1-dev", "Kolors"])
+```
+
+Download your own models.
+
+```python
+from diffsynth.models.downloader import download_from_huggingface, download_from_modelscope
+
+# From Modelscope (recommended)
+download_from_modelscope("Kwai-Kolors/Kolors", "vae/diffusion_pytorch_model.fp16.bin", "models/kolors/Kolors/vae")
+# From Huggingface
+download_from_huggingface("Kwai-Kolors/Kolors", "vae/diffusion_pytorch_model.fp16.safetensors", "models/kolors/Kolors/vae")
+```
+
+### Video Synthesis
+
+#### Text-to-video using CogVideoX-5B
+
+CogVideoX-5B is released by ZhiPu. We provide an improved pipeline, supporting text-to-video, video editing, self-upscaling and video interpolation. [`examples/video_synthesis`](./examples/video_synthesis/)
+
+The video on the left is generated using the original text-to-video pipeline, while the video on the right is the result after editing and frame interpolation.
+
+https://github.com/user-attachments/assets/26b044c1-4a60-44a4-842f-627ff289d006
+
+#### Long Video Synthesis
+
+We trained extended video synthesis models, which can generate 128 frames. [`examples/ExVideo`](./examples/ExVideo/)
+
+https://github.com/modelscope/DiffSynth-Studio/assets/35051019/d97f6aa9-8064-4b5b-9d49-ed6001bb9acc
+
+https://github.com/user-attachments/assets/321ee04b-8c17-479e-8a95-8cbcf21f8d7e
+
+#### Toon Shading
+
+Render realistic videos in a flatten style and enable video editing features. [`examples/Diffutoon`](./examples/Diffutoon/)
+
+https://github.com/Artiprocher/DiffSynth-Studio/assets/35051019/b54c05c5-d747-4709-be5e-b39af82404dd
+
+https://github.com/Artiprocher/DiffSynth-Studio/assets/35051019/20528af5-5100-474a-8cdc-440b9efdd86c
+
+#### Video Stylization
+
+Video stylization without video models. [`examples/diffsynth`](./examples/diffsynth/)
+
+https://github.com/Artiprocher/DiffSynth-Studio/assets/35051019/59fb2f7b-8de0-4481-b79f-0c3a7361a1ea
+
+### Image Synthesis
+
+Generate high-resolution images, by breaking the limitation of diffusion models! [`examples/image_synthesis`](./examples/image_synthesis/).
+
+LoRA fine-tuning is supported in [`examples/train`](./examples/train/).
+
+|FLUX|Stable Diffusion 3|
+|-|-|
+|![image_1024_cfg](https://github.com/user-attachments/assets/984561e9-553d-4952-9443-79ce144f379f)|![image_1024](https://github.com/modelscope/DiffSynth-Studio/assets/35051019/4df346db-6f91-420a-b4c1-26e205376098)|
+
+|Kolors|Hunyuan-DiT|
+|-|-|
+|![image_1024](https://github.com/modelscope/DiffSynth-Studio/assets/35051019/53ef6f41-da11-4701-8665-9f64392607bf)|![image_1024](https://github.com/modelscope/DiffSynth-Studio/assets/35051019/60b022c8-df3f-4541-95ab-bf39f2fa8bb5)|
+
+|Stable Diffusion|Stable Diffusion XL|
+|-|-|
+|![1024](https://github.com/Artiprocher/DiffSynth-Studio/assets/35051019/6fc84611-8da6-4a1f-8fee-9a34eba3b4a5)|![1024](https://github.com/Artiprocher/DiffSynth-Studio/assets/35051019/67687748-e738-438c-aee5-96096f09ac90)|
+
+## Usage (in WebUI)
+
+Create stunning images using the painter, with assistance from AI!
+
+https://github.com/user-attachments/assets/95265d21-cdd6-4125-a7cb-9fbcf6ceb7b0
+
+**This video is not rendered in real-time.**
+
+Before launching the WebUI, please download models to the folder `./models`. See [here](#download-models).
+
+* `Gradio` version
+
+```
+pip install gradio
+```
+
+```
+python apps/gradio/DiffSynth_Studio.py
+```
+
+![20240822102002](https://github.com/user-attachments/assets/59613157-de51-4109-99b3-97cbffd88076)
+
+* `Streamlit` version
+
+```
+pip install streamlit streamlit-drawable-canvas
+```
+
+```
+python -m streamlit run apps/streamlit/DiffSynth_Studio.py
+```
+
+https://github.com/Artiprocher/DiffSynth-Studio/assets/35051019/93085557-73f3-4eee-a205-9829591ef954

apps/gradio/DiffSynth_Studio.py

@@ -0,0 +1,252 @@
+import gradio as gr
+from diffsynth import ModelManager, SDImagePipeline, SDXLImagePipeline, SD3ImagePipeline, HunyuanDiTImagePipeline, FluxImagePipeline
+import os, torch
+from PIL import Image
+import numpy as np
+
+
+config = {
+    "model_config": {
+        "Stable Diffusion": {
+            "model_folder": "models/stable_diffusion",
+            "pipeline_class": SDImagePipeline,
+            "default_parameters": {
+                "cfg_scale": 7.0,
+                "height": 512,
+                "width": 512,
+            }
+        },
+        "Stable Diffusion XL": {
+            "model_folder": "models/stable_diffusion_xl",
+            "pipeline_class": SDXLImagePipeline,
+            "default_parameters": {
+                "cfg_scale": 7.0,
+            }
+        },
+        "Stable Diffusion 3": {
+            "model_folder": "models/stable_diffusion_3",
+            "pipeline_class": SD3ImagePipeline,
+            "default_parameters": {
+                "cfg_scale": 7.0,
+            }
+        },
+        "Stable Diffusion XL Turbo": {
+            "model_folder": "models/stable_diffusion_xl_turbo",
+            "pipeline_class": SDXLImagePipeline,
+            "default_parameters": {
+                "negative_prompt": "",
+                "cfg_scale": 1.0,
+                "num_inference_steps": 1,
+                "height": 512,
+                "width": 512,
+            }
+        },
+        "Kolors": {
+            "model_folder": "models/kolors",
+            "pipeline_class": SDXLImagePipeline,
+            "default_parameters": {
+                "cfg_scale": 7.0,
+            }
+        },
+        "HunyuanDiT": {
+            "model_folder": "models/HunyuanDiT",
+            "pipeline_class": HunyuanDiTImagePipeline,
+            "default_parameters": {
+                "cfg_scale": 7.0,
+            }
+        },
+        "FLUX": {
+            "model_folder": "models/FLUX",
+            "pipeline_class": FluxImagePipeline,
+            "default_parameters": {
+                "cfg_scale": 1.0,
+            }
+        }
+    },
+    "max_num_painter_layers": 8,
+    "max_num_model_cache": 1,
+}
+
+
+def load_model_list(model_type):
+    if model_type is None:
+        return []
+    folder = config["model_config"][model_type]["model_folder"]
+    file_list = [i for i in os.listdir(folder) if i.endswith(".safetensors")]
+    if model_type in ["HunyuanDiT", "Kolors", "FLUX"]:
+        file_list += [i for i in os.listdir(folder) if os.path.isdir(os.path.join(folder, i))]
+    file_list = sorted(file_list)
+    return file_list
+
+
+def load_model(model_type, model_path):
+    global model_dict
+    model_key = f"{model_type}:{model_path}"
+    if model_key in model_dict:
+        return model_dict[model_key]
+    model_path = os.path.join(config["model_config"][model_type]["model_folder"], model_path)
+    model_manager = ModelManager()
+    if model_type == "HunyuanDiT":
+        model_manager.load_models([
+            os.path.join(model_path, "clip_text_encoder/pytorch_model.bin"),
+            os.path.join(model_path, "mt5/pytorch_model.bin"),
+            os.path.join(model_path, "model/pytorch_model_ema.pt"),
+            os.path.join(model_path, "sdxl-vae-fp16-fix/diffusion_pytorch_model.bin"),
+        ])
+    elif model_type == "Kolors":
+        model_manager.load_models([
+            os.path.join(model_path, "text_encoder"),
+            os.path.join(model_path, "unet/diffusion_pytorch_model.safetensors"),
+            os.path.join(model_path, "vae/diffusion_pytorch_model.safetensors"),
+        ])
+    elif model_type == "FLUX":
+        model_manager.torch_dtype = torch.bfloat16
+        file_list = [
+            os.path.join(model_path, "text_encoder/model.safetensors"),
+            os.path.join(model_path, "text_encoder_2"),
+        ]
+        for file_name in os.listdir(model_path):
+            if file_name.endswith(".safetensors"):
+                file_list.append(os.path.join(model_path, file_name))
+        model_manager.load_models(file_list)
+    else:
+        model_manager.load_model(model_path)
+    pipe = config["model_config"][model_type]["pipeline_class"].from_model_manager(model_manager)
+    while len(model_dict) + 1 > config["max_num_model_cache"]:
+        key = next(iter(model_dict.keys()))
+        model_manager_to_release, _ = model_dict[key]
+        model_manager_to_release.to("cpu")
+        del model_dict[key]
+        torch.cuda.empty_cache()
+    model_dict[model_key] = model_manager, pipe
+    return model_manager, pipe
+
+
+model_dict = {}
+
+with gr.Blocks() as app:
+    gr.Markdown("# DiffSynth-Studio Painter")
+    with gr.Row():
+        with gr.Column(scale=382, min_width=100):
+
+            with gr.Accordion(label="Model"):
+                model_type = gr.Dropdown(choices=[i for i in config["model_config"]], label="Model type")
+                model_path = gr.Dropdown(choices=[], interactive=True, label="Model path")
+
+                @gr.on(inputs=model_type, outputs=model_path, triggers=model_type.change)
+                def model_type_to_model_path(model_type):
+                    return gr.Dropdown(choices=load_model_list(model_type))
+                
+            with gr.Accordion(label="Prompt"):
+                prompt = gr.Textbox(label="Prompt", lines=3)
+                negative_prompt = gr.Textbox(label="Negative prompt", lines=1)
+                cfg_scale = gr.Slider(minimum=1.0, maximum=10.0, value=7.0, step=0.1, interactive=True, label="Classifier-free guidance scale")
+                embedded_guidance = gr.Slider(minimum=0.0, maximum=10.0, value=0.0, step=0.1, interactive=True, label="Embedded guidance scale (only for FLUX)")
+            
+            with gr.Accordion(label="Image"):
+                num_inference_steps = gr.Slider(minimum=1, maximum=100, value=20, step=1, interactive=True, label="Inference steps")
+                height = gr.Slider(minimum=64, maximum=2048, value=1024, step=64, interactive=True, label="Height")
+                width = gr.Slider(minimum=64, maximum=2048, value=1024, step=64, interactive=True, label="Width")
+                with gr.Column():
+                    use_fixed_seed = gr.Checkbox(value=True, interactive=False, label="Use fixed seed")
+                    seed = gr.Number(minimum=0, maximum=10**9, value=0, interactive=True, label="Random seed", show_label=False)
+
+            @gr.on(
+                inputs=[model_type, model_path, prompt, negative_prompt, cfg_scale, embedded_guidance, num_inference_steps, height, width],
+                outputs=[prompt, negative_prompt, cfg_scale, embedded_guidance, num_inference_steps, height, width],
+                triggers=model_path.change
+            )
+            def model_path_to_default_params(model_type, model_path, prompt, negative_prompt, cfg_scale, embedded_guidance, num_inference_steps, height, width):
+                load_model(model_type, model_path)
+                cfg_scale = config["model_config"][model_type]["default_parameters"].get("cfg_scale", cfg_scale)
+                embedded_guidance = config["model_config"][model_type]["default_parameters"].get("embedded_guidance", embedded_guidance)
+                num_inference_steps = config["model_config"][model_type]["default_parameters"].get("num_inference_steps", num_inference_steps)
+                height = config["model_config"][model_type]["default_parameters"].get("height", height)
+                width = config["model_config"][model_type]["default_parameters"].get("width", width)
+                return prompt, negative_prompt, cfg_scale, embedded_guidance, num_inference_steps, height, width
+                
+
+        with gr.Column(scale=618, min_width=100):
+            with gr.Accordion(label="Painter"):
+                enable_local_prompt_list = []
+                local_prompt_list = []
+                mask_scale_list = []
+                canvas_list = []
+                for painter_layer_id in range(config["max_num_painter_layers"]):
+                    with gr.Tab(label=f"Layer {painter_layer_id}"):
+                        enable_local_prompt = gr.Checkbox(label="Enable", value=False, key=f"enable_local_prompt_{painter_layer_id}")
+                        local_prompt = gr.Textbox(label="Local prompt", key=f"local_prompt_{painter_layer_id}")
+                        mask_scale = gr.Slider(minimum=0.0, maximum=5.0, value=1.0, step=0.1, interactive=True, label="Mask scale", key=f"mask_scale_{painter_layer_id}")
+                        canvas = gr.ImageEditor(canvas_size=(512, 1), sources=None, layers=False, interactive=True, image_mode="RGBA",
+                                                brush=gr.Brush(default_size=100, default_color="#000000", colors=["#000000"]),
+                                                label="Painter", key=f"canvas_{painter_layer_id}")
+                        @gr.on(inputs=[height, width, canvas], outputs=canvas, triggers=[height.change, width.change, canvas.clear, enable_local_prompt.change], show_progress="hidden")
+                        def resize_canvas(height, width, canvas):
+                            h, w = canvas["background"].shape[:2]
+                            if h != height or width != w:
+                                return np.ones((height, width, 3), dtype=np.uint8) * 255
+                            else:
+                                return canvas
+                        
+                        enable_local_prompt_list.append(enable_local_prompt)
+                        local_prompt_list.append(local_prompt)
+                        mask_scale_list.append(mask_scale)
+                        canvas_list.append(canvas)
+            with gr.Accordion(label="Results"):
+                run_button = gr.Button(value="Generate", variant="primary")
+                output_image = gr.Image(sources=None, show_label=False, interactive=False, type="pil")
+                with gr.Row():
+                    with gr.Column():
+                        output_to_painter_button = gr.Button(value="Set as painter's background")
+                    with gr.Column():
+                        output_to_input_button = gr.Button(value="Set as input image")
+                painter_background = gr.State(None)
+                input_background = gr.State(None)
+                @gr.on(
+                    inputs=[model_type, model_path, prompt, negative_prompt, cfg_scale, embedded_guidance, num_inference_steps, height, width, seed] + enable_local_prompt_list + local_prompt_list + mask_scale_list + canvas_list,
+                    outputs=[output_image],
+                    triggers=run_button.click
+                )
+                def generate_image(model_type, model_path, prompt, negative_prompt, cfg_scale, embedded_guidance, num_inference_steps, height, width, seed, *args, progress=gr.Progress()):
+                    _, pipe = load_model(model_type, model_path)
+                    input_params = {
+                        "prompt": prompt,
+                        "negative_prompt": negative_prompt,
+                        "cfg_scale": cfg_scale,
+                        "num_inference_steps": num_inference_steps,
+                        "height": height,
+                        "width": width,
+                        "progress_bar_cmd": progress.tqdm,
+                    }
+                    if isinstance(pipe, FluxImagePipeline):
+                        input_params["embedded_guidance"] = embedded_guidance
+                    enable_local_prompt_list, local_prompt_list, mask_scale_list, canvas_list = (
+                        args[0 * config["max_num_painter_layers"]: 1 * config["max_num_painter_layers"]],
+                        args[1 * config["max_num_painter_layers"]: 2 * config["max_num_painter_layers"]],
+                        args[2 * config["max_num_painter_layers"]: 3 * config["max_num_painter_layers"]],
+                        args[3 * config["max_num_painter_layers"]: 4 * config["max_num_painter_layers"]]
+                    )
+                    local_prompts, masks, mask_scales = [], [], []
+                    for enable_local_prompt, local_prompt, mask_scale, canvas in zip(
+                        enable_local_prompt_list, local_prompt_list, mask_scale_list, canvas_list
+                    ):
+                        if enable_local_prompt:
+                            local_prompts.append(local_prompt)
+                            masks.append(Image.fromarray(canvas["layers"][0][:, :, -1]).convert("RGB"))
+                            mask_scales.append(mask_scale)
+                    input_params.update({
+                        "local_prompts": local_prompts,
+                        "masks": masks,
+                        "mask_scales": mask_scales,
+                    })
+                    torch.manual_seed(seed)
+                    image = pipe(**input_params)
+                    return image
+                
+                @gr.on(inputs=[output_image] + canvas_list, outputs=canvas_list, triggers=output_to_painter_button.click)
+                def send_output_to_painter_background(output_image, *canvas_list):
+                    for canvas in canvas_list:
+                        h, w = canvas["background"].shape[:2]
+                        canvas["background"] = output_image.resize((w, h))
+                    return tuple(canvas_list)
+app.launch()

apps/gradio/entity_level_control.py

@@ -0,0 +1,390 @@
+import os
+import torch
+import numpy as np
+from PIL import Image, ImageDraw, ImageFont
+import random
+import json
+import gradio as gr
+from diffsynth import ModelManager, FluxImagePipeline, download_customized_models
+from modelscope import dataset_snapshot_download
+
+
+dataset_snapshot_download(dataset_id="DiffSynth-Studio/examples_in_diffsynth", local_dir="./", allow_file_pattern=f"data/examples/eligen/entity_control/*")
+example_json = 'data/examples/eligen/entity_control/ui_examples.json'
+with open(example_json, 'r') as f:
+    examples = json.load(f)['examples']
+
+for idx in range(len(examples)):
+    example_id = examples[idx]['example_id']
+    entity_prompts = examples[idx]['local_prompt_list']
+    examples[idx]['mask_lists'] = [Image.open(f"data/examples/eligen/entity_control/example_{example_id}/{i}.png").convert('RGB') for i in range(len(entity_prompts))]
+
+def create_canvas_data(background, masks):
+    if background.shape[-1] == 3:
+        background = np.dstack([background, np.full(background.shape[:2], 255, dtype=np.uint8)])
+    layers = []
+    for mask in masks:
+        if mask is not None:
+            mask_single_channel = mask if mask.ndim == 2 else mask[..., 0]
+            layer = np.zeros((mask_single_channel.shape[0], mask_single_channel.shape[1], 4), dtype=np.uint8)
+            layer[..., -1] = mask_single_channel
+            layers.append(layer)
+        else:
+            layers.append(np.zeros_like(background))
+
+    composite = background.copy()
+    for layer in layers:
+        if layer.size > 0:
+            composite = np.where(layer[..., -1:] > 0, layer, composite)
+    return {
+        "background": background,
+        "layers": layers,
+        "composite": composite,
+    }
+
+def load_example(load_example_button):
+    example_idx = int(load_example_button.split()[-1]) - 1
+    example = examples[example_idx]
+    result = [
+        50,
+        example["global_prompt"],
+        example["negative_prompt"],
+        example["seed"],
+        *example["local_prompt_list"],
+    ]
+    num_entities = len(example["local_prompt_list"])
+    result += [""] * (config["max_num_painter_layers"] - num_entities)
+    masks = []
+    for mask in example["mask_lists"]:
+        mask_single_channel = np.array(mask.convert("L"))
+        masks.append(mask_single_channel)
+    for _ in range(config["max_num_painter_layers"] - len(masks)):
+        blank_mask = np.zeros_like(masks[0]) if masks else np.zeros((512, 512), dtype=np.uint8)
+        masks.append(blank_mask)
+    background = np.ones((masks[0].shape[0], masks[0].shape[1], 4), dtype=np.uint8) * 255
+    canvas_data_list = []
+    for mask in masks:
+        canvas_data = create_canvas_data(background, [mask])
+        canvas_data_list.append(canvas_data)
+    result.extend(canvas_data_list)
+    return result
+
+def save_mask_prompts(masks, mask_prompts, global_prompt, seed=0, random_dir='0000000'):
+    save_dir = os.path.join('workdirs/tmp_mask', random_dir)
+    print(f'save to {save_dir}')
+    os.makedirs(save_dir, exist_ok=True)
+    for i, mask in enumerate(masks):
+        save_path = os.path.join(save_dir, f'{i}.png')
+        mask.save(save_path)
+    sample = {
+        "global_prompt": global_prompt,
+        "mask_prompts": mask_prompts,
+        "seed": seed,
+    }
+    with open(os.path.join(save_dir, f"prompts.json"), 'w') as f:
+        json.dump(sample, f, indent=4)
+
+def visualize_masks(image, masks, mask_prompts, font_size=35, use_random_colors=False):
+    # Create a blank image for overlays
+    overlay = Image.new('RGBA', image.size, (0, 0, 0, 0))
+    colors = [
+        (165, 238, 173, 80),
+        (76, 102, 221, 80),
+        (221, 160, 77, 80),
+        (204, 93, 71, 80),
+        (145, 187, 149, 80),
+        (134, 141, 172, 80),
+        (157, 137, 109, 80),
+        (153, 104, 95, 80),
+        (165, 238, 173, 80),
+        (76, 102, 221, 80),
+        (221, 160, 77, 80),
+        (204, 93, 71, 80),
+        (145, 187, 149, 80),
+        (134, 141, 172, 80),
+        (157, 137, 109, 80),
+        (153, 104, 95, 80),
+    ]
+    # Generate random colors for each mask
+    if use_random_colors:
+        colors = [(random.randint(0, 255), random.randint(0, 255), random.randint(0, 255), 80) for _ in range(len(masks))]
+    # Font settings
+    try:
+        font = ImageFont.truetype("arial", font_size)  # Adjust as needed
+    except IOError:
+        font = ImageFont.load_default(font_size)
+    # Overlay each mask onto the overlay image
+    for mask, mask_prompt, color in zip(masks, mask_prompts, colors):
+        if mask is None:
+            continue
+        # Convert mask to RGBA mode
+        mask_rgba = mask.convert('RGBA')
+        mask_data = mask_rgba.getdata()
+        new_data = [(color if item[:3] == (255, 255, 255) else (0, 0, 0, 0)) for item in mask_data]
+        mask_rgba.putdata(new_data)
+        # Draw the mask prompt text on the mask
+        draw = ImageDraw.Draw(mask_rgba)
+        mask_bbox = mask.getbbox()  # Get the bounding box of the mask
+        if mask_bbox is None:
+            continue
+        text_position = (mask_bbox[0] + 10, mask_bbox[1] + 10)  # Adjust text position based on mask position
+        draw.text(text_position, mask_prompt, fill=(255, 255, 255, 255), font=font)
+        # Alpha composite the overlay with this mask
+        overlay = Image.alpha_composite(overlay, mask_rgba)
+    # Composite the overlay onto the original image
+    result = Image.alpha_composite(image.convert('RGBA'), overlay)
+    return result
+
+config = {
+    "model_config": {
+        "FLUX": {
+            "model_folder": "models/FLUX",
+            "pipeline_class": FluxImagePipeline,
+            "default_parameters": {
+                "cfg_scale": 3.0,
+                "embedded_guidance": 3.5,
+                "num_inference_steps": 30,
+            }
+        },
+    },
+    "max_num_painter_layers": 8,
+    "max_num_model_cache": 1,
+}
+
+model_dict = {}
+
+def load_model(model_type='FLUX', model_path='FLUX.1-dev'):
+    global model_dict
+    model_key = f"{model_type}:{model_path}"
+    if model_key in model_dict:
+        return model_dict[model_key]
+    model_path = os.path.join(config["model_config"][model_type]["model_folder"], model_path)
+    model_manager = ModelManager(torch_dtype=torch.bfloat16, device="cuda", model_id_list=["FLUX.1-dev"])
+    model_manager.load_lora(
+        download_customized_models(
+            model_id="DiffSynth-Studio/Eligen",
+            origin_file_path="model_bf16.safetensors",
+            local_dir="models/lora/entity_control",
+        ),
+        lora_alpha=1,
+    )
+    pipe = config["model_config"][model_type]["pipeline_class"].from_model_manager(model_manager)
+    model_dict[model_key] = model_manager, pipe
+    return model_manager, pipe
+
+
+with gr.Blocks() as app:
+    gr.Markdown(
+        """## EliGen: Entity-Level Controllable Text-to-Image Model
+                1. On the left, input the **global prompt** for the overall image, such as "a person stands by the river."
+                2. On the right, input the **local prompt** for each entity, such as "person," and draw the corresponding mask in the **Entity Mask Painter**. Generally, solid rectangular masks yield better results.
+                3. Click the **Generate** button to create the image. By selecting different **random seeds**, you can generate diverse images.
+                4. **You can directly click the "Load Example" button on any sample at the bottom to load example inputs.**
+                """
+    )
+
+    loading_status = gr.Textbox(label="Loading Model...", value="Loading model... Please wait...", visible=True)
+    main_interface = gr.Column(visible=False)
+
+    def initialize_model():
+        try:
+            load_model()
+            return {
+                loading_status: gr.update(value="Model loaded successfully!", visible=False),
+                main_interface: gr.update(visible=True),
+            }
+        except Exception as e:
+            print(f'Failed to load model with error: {e}')
+            return {
+                loading_status: gr.update(value=f"Failed to load model: {str(e)}", visible=True),
+                main_interface: gr.update(visible=True),
+            }
+
+    app.load(initialize_model, inputs=None, outputs=[loading_status, main_interface])
+
+    with main_interface:
+        with gr.Row():
+            local_prompt_list = []
+            canvas_list = []
+            random_mask_dir = gr.State(f'{random.randint(0, 1000000):08d}')
+            with gr.Column(scale=382, min_width=100):
+                model_type = gr.State('FLUX')
+                model_path = gr.State('FLUX.1-dev')
+                with gr.Accordion(label="Global prompt"):
+                    prompt = gr.Textbox(label="Global Prompt", lines=3)
+                    negative_prompt = gr.Textbox(label="Negative prompt", value="worst quality, low quality, monochrome, zombie, interlocked fingers, Aissist, cleavage, nsfw, blur,", lines=3)
+                with gr.Accordion(label="Inference Options", open=True):
+                    seed = gr.Number(minimum=0, maximum=10**9, value=42, interactive=True, label="Random seed", show_label=True)
+                    num_inference_steps = gr.Slider(minimum=1, maximum=100, value=30, step=1, interactive=True, label="Inference steps")
+                    cfg_scale = gr.Slider(minimum=2.0, maximum=10.0, value=3.0, step=0.1, interactive=True, label="Classifier-free guidance scale")
+                    embedded_guidance = gr.Slider(minimum=0.0, maximum=10.0, value=3.5, step=0.1, interactive=True, label="Embedded guidance scale")
+                    height = gr.Slider(minimum=64, maximum=2048, value=1024, step=64, interactive=True, label="Height")
+                    width = gr.Slider(minimum=64, maximum=2048, value=1024, step=64, interactive=True, label="Width")
+                with gr.Accordion(label="Inpaint Input Image", open=False):
+                    input_image = gr.Image(sources=None, show_label=False, interactive=True, type="pil")
+                    background_weight = gr.Slider(minimum=0.0, maximum=1000., value=0., step=1, interactive=False, label="background_weight", visible=False)
+
+                    with gr.Column():
+                        reset_input_button = gr.Button(value="Reset Inpaint Input")
+                        send_input_to_painter = gr.Button(value="Set as painter's background")
+                    @gr.on(inputs=[input_image], outputs=[input_image], triggers=reset_input_button.click)
+                    def reset_input_image(input_image):
+                        return None
+
+            with gr.Column(scale=618, min_width=100):
+                with gr.Accordion(label="Entity Painter"):
+                    for painter_layer_id in range(config["max_num_painter_layers"]):
+                        with gr.Tab(label=f"Entity {painter_layer_id}"):
+                            local_prompt = gr.Textbox(label="Local prompt", key=f"local_prompt_{painter_layer_id}")
+                            canvas = gr.ImageEditor(
+                                canvas_size=(512, 512),
+                                sources=None,
+                                layers=False,
+                                interactive=True,
+                                image_mode="RGBA",
+                                brush=gr.Brush(
+                                    default_size=50,
+                                    default_color="#000000",
+                                    colors=["#000000"],
+                                ),
+                                label="Entity Mask Painter",
+                                key=f"canvas_{painter_layer_id}",
+                                width=width,
+                                height=height,
+                            )
+                            @gr.on(inputs=[height, width, canvas], outputs=canvas, triggers=[height.change, width.change, canvas.clear], show_progress="hidden")
+                            def resize_canvas(height, width, canvas):
+                                h, w = canvas["background"].shape[:2]
+                                if h != height or width != w:
+                                    return np.ones((height, width, 3), dtype=np.uint8) * 255
+                                else:
+                                    return canvas
+                            local_prompt_list.append(local_prompt)
+                            canvas_list.append(canvas)
+                with gr.Accordion(label="Results"):
+                    run_button = gr.Button(value="Generate", variant="primary")
+                    output_image = gr.Image(sources=None, show_label=False, interactive=False, type="pil")
+                    with gr.Row():
+                        with gr.Column():
+                            output_to_painter_button = gr.Button(value="Set as painter's background")
+                        with gr.Column():
+                            return_with_mask = gr.Checkbox(value=False, interactive=True, label="show result with mask painting")
+                            output_to_input_button = gr.Button(value="Set as input image", visible=False, interactive=False)
+                    real_output = gr.State(None)
+                    mask_out = gr.State(None)
+
+                    @gr.on(
+                        inputs=[model_type, model_path, prompt, negative_prompt, cfg_scale, embedded_guidance, num_inference_steps, height, width, return_with_mask, seed, input_image, background_weight, random_mask_dir] + local_prompt_list + canvas_list,
+                        outputs=[output_image, real_output, mask_out],
+                        triggers=run_button.click
+                    )
+                    def generate_image(model_type, model_path, prompt, negative_prompt, cfg_scale, embedded_guidance, num_inference_steps, height, width, return_with_mask, seed, input_image, background_weight, random_mask_dir, *args, progress=gr.Progress()):
+                        _, pipe = load_model(model_type, model_path)
+                        input_params = {
+                            "prompt": prompt,
+                            "negative_prompt": negative_prompt,
+                            "cfg_scale": cfg_scale,
+                            "num_inference_steps": num_inference_steps,
+                            "height": height,
+                            "width": width,
+                            "progress_bar_cmd": progress.tqdm,
+                        }
+                        if isinstance(pipe, FluxImagePipeline):
+                            input_params["embedded_guidance"] = embedded_guidance
+                        if input_image is not None:
+                            input_params["input_image"] = input_image.resize((width, height)).convert("RGB")
+                            input_params["enable_eligen_inpaint"] = True
+
+                        local_prompt_list, canvas_list = (
+                            args[0 * config["max_num_painter_layers"]: 1 * config["max_num_painter_layers"]],
+                            args[1 * config["max_num_painter_layers"]: 2 * config["max_num_painter_layers"]],
+                        )
+                        local_prompts, masks = [], []
+                        for local_prompt, canvas in zip(local_prompt_list, canvas_list):
+                            if isinstance(local_prompt, str) and len(local_prompt) > 0:
+                                local_prompts.append(local_prompt)
+                                masks.append(Image.fromarray(canvas["layers"][0][:, :, -1]).convert("RGB"))
+                        entity_masks = None if len(masks) == 0 else masks
+                        entity_prompts = None if len(local_prompts) == 0 else local_prompts
+                        input_params.update({
+                            "eligen_entity_prompts": entity_prompts,
+                            "eligen_entity_masks": entity_masks,
+                        })
+                        torch.manual_seed(seed)
+                        # save_mask_prompts(masks, local_prompts, prompt, seed, random_mask_dir)
+                        image = pipe(**input_params)
+                        masks = [mask.resize(image.size) for mask in masks]
+                        image_with_mask = visualize_masks(image, masks, local_prompts)
+
+                        real_output = gr.State(image)
+                        mask_out = gr.State(image_with_mask)
+
+                        if return_with_mask:
+                            return image_with_mask, real_output, mask_out
+                        return image, real_output, mask_out
+
+                    @gr.on(inputs=[input_image] + canvas_list, outputs=canvas_list, triggers=send_input_to_painter.click)
+                    def send_input_to_painter_background(input_image, *canvas_list):
+                        if input_image is None:
+                            return tuple(canvas_list)
+                        for canvas in canvas_list:
+                            h, w = canvas["background"].shape[:2]
+                            canvas["background"] = input_image.resize((w, h))
+                        return tuple(canvas_list)
+                    @gr.on(inputs=[real_output] + canvas_list, outputs=canvas_list, triggers=output_to_painter_button.click)
+                    def send_output_to_painter_background(real_output, *canvas_list):
+                        if real_output is None:
+                            return tuple(canvas_list)
+                        for canvas in canvas_list:
+                            h, w = canvas["background"].shape[:2]
+                            canvas["background"] = real_output.value.resize((w, h))
+                        return tuple(canvas_list)
+                    @gr.on(inputs=[return_with_mask, real_output, mask_out], outputs=[output_image], triggers=[return_with_mask.change], show_progress="hidden")
+                    def show_output(return_with_mask, real_output, mask_out):
+                        if return_with_mask:
+                            return mask_out.value
+                        else:
+                            return real_output.value
+                    @gr.on(inputs=[real_output], outputs=[input_image], triggers=output_to_input_button.click)
+                    def send_output_to_pipe_input(real_output):
+                        return real_output.value
+
+        with gr.Column():
+            gr.Markdown("## Examples")
+            for i in range(0, len(examples), 2):
+                with gr.Row():
+                    if i < len(examples):
+                        example = examples[i]
+                        with gr.Column():
+                            example_image = gr.Image(
+                                value=f"data/examples/eligen/entity_control/example_{example['example_id']}/example_image.png",
+                                label=example["description"],
+                                interactive=False,
+                                width=1024,
+                                height=512
+                            )
+                            load_example_button = gr.Button(value=f"Load Example {example['example_id']}")
+                            load_example_button.click(
+                                load_example,
+                                inputs=[load_example_button],
+                                outputs=[num_inference_steps, prompt, negative_prompt, seed] + local_prompt_list + canvas_list
+                            )
+
+                    if i + 1 < len(examples):
+                        example = examples[i + 1]
+                        with gr.Column():
+                            example_image = gr.Image(
+                                value=f"data/examples/eligen/entity_control/example_{example['example_id']}/example_image.png",
+                                label=example["description"],
+                                interactive=False,
+                                width=1024,
+                                height=512
+                            )
+                            load_example_button = gr.Button(value=f"Load Example {example['example_id']}")
+                            load_example_button.click(
+                                load_example,
+                                inputs=[load_example_button],
+                                outputs=[num_inference_steps, prompt, negative_prompt, seed] + local_prompt_list + canvas_list
+                            )
+app.config["show_progress"] = "hidden"
+app.launch()

apps/streamlit/DiffSynth_Studio.py

@@ -0,0 +1,15 @@
+# Set web page format
+import streamlit as st
+st.set_page_config(layout="wide")
+# Disable virtual VRAM on windows system
+import torch
+torch.cuda.set_per_process_memory_fraction(0.999, 0)
+
+
+st.markdown("""
+# DiffSynth Studio
+
+[Source Code](https://github.com/Artiprocher/DiffSynth-Studio)
+
+Welcome to DiffSynth Studio.
+""")

apps/streamlit/pages/1_Image_Creator.py

@@ -0,0 +1,362 @@
+import torch, os, io, json, time
+import numpy as np
+from PIL import Image
+import streamlit as st
+st.set_page_config(layout="wide")
+from streamlit_drawable_canvas import st_canvas
+from diffsynth.models import ModelManager
+from diffsynth.pipelines import SDImagePipeline, SDXLImagePipeline, SD3ImagePipeline, HunyuanDiTImagePipeline, FluxImagePipeline
+from diffsynth.data.video import crop_and_resize
+
+
+config = {
+    "Stable Diffusion": {
+        "model_folder": "models/stable_diffusion",
+        "pipeline_class": SDImagePipeline,
+        "fixed_parameters": {}
+    },
+    "Stable Diffusion XL": {
+        "model_folder": "models/stable_diffusion_xl",
+        "pipeline_class": SDXLImagePipeline,
+        "fixed_parameters": {}
+    },
+    "Stable Diffusion 3": {
+        "model_folder": "models/stable_diffusion_3",
+        "pipeline_class": SD3ImagePipeline,
+        "fixed_parameters": {}
+    },
+    "Stable Diffusion XL Turbo": {
+        "model_folder": "models/stable_diffusion_xl_turbo",
+        "pipeline_class": SDXLImagePipeline,
+        "fixed_parameters": {
+            "negative_prompt": "",
+            "cfg_scale": 1.0,
+            "num_inference_steps": 1,
+            "height": 512,
+            "width": 512,
+        }
+    },
+    "Kolors": {
+        "model_folder": "models/kolors",
+        "pipeline_class": SDXLImagePipeline,
+        "fixed_parameters": {}
+    },
+    "HunyuanDiT": {
+        "model_folder": "models/HunyuanDiT",
+        "pipeline_class": HunyuanDiTImagePipeline,
+        "fixed_parameters": {
+            "height": 1024,
+            "width": 1024,
+        }
+    },
+    "FLUX": {
+        "model_folder": "models/FLUX",
+        "pipeline_class": FluxImagePipeline,
+        "fixed_parameters": {
+            "cfg_scale": 1.0,
+        }
+    }
+}
+
+
+def load_model_list(model_type):
+    folder = config[model_type]["model_folder"]
+    file_list = [i for i in os.listdir(folder) if i.endswith(".safetensors")]
+    if model_type in ["HunyuanDiT", "Kolors", "FLUX"]:
+        file_list += [i for i in os.listdir(folder) if os.path.isdir(os.path.join(folder, i))]
+    file_list = sorted(file_list)
+    return file_list
+
+
+def release_model():
+    if "model_manager" in st.session_state:
+        st.session_state["model_manager"].to("cpu")
+        del st.session_state["loaded_model_path"]
+        del st.session_state["model_manager"]
+        del st.session_state["pipeline"]
+        torch.cuda.empty_cache()
+
+
+def load_model(model_type, model_path):
+    model_manager = ModelManager()
+    if model_type == "HunyuanDiT":
+        model_manager.load_models([
+            os.path.join(model_path, "clip_text_encoder/pytorch_model.bin"),
+            os.path.join(model_path, "mt5/pytorch_model.bin"),
+            os.path.join(model_path, "model/pytorch_model_ema.pt"),
+            os.path.join(model_path, "sdxl-vae-fp16-fix/diffusion_pytorch_model.bin"),
+        ])
+    elif model_type == "Kolors":
+        model_manager.load_models([
+            os.path.join(model_path, "text_encoder"),
+            os.path.join(model_path, "unet/diffusion_pytorch_model.safetensors"),
+            os.path.join(model_path, "vae/diffusion_pytorch_model.safetensors"),
+        ])
+    elif model_type == "FLUX":
+        model_manager.torch_dtype = torch.bfloat16
+        file_list = [
+            os.path.join(model_path, "text_encoder/model.safetensors"),
+            os.path.join(model_path, "text_encoder_2"),
+        ]
+        for file_name in os.listdir(model_path):
+            if file_name.endswith(".safetensors"):
+                file_list.append(os.path.join(model_path, file_name))
+        model_manager.load_models(file_list)
+    else:
+        model_manager.load_model(model_path)
+    pipeline = config[model_type]["pipeline_class"].from_model_manager(model_manager)
+    st.session_state.loaded_model_path = model_path
+    st.session_state.model_manager = model_manager
+    st.session_state.pipeline = pipeline
+    return model_manager, pipeline
+
+
+def use_output_image_as_input(update=True):
+    # Search for input image
+    output_image_id = 0
+    selected_output_image = None
+    while True:
+        if f"use_output_as_input_{output_image_id}" not in st.session_state:
+            break
+        if st.session_state[f"use_output_as_input_{output_image_id}"]:
+            selected_output_image = st.session_state["output_images"][output_image_id]
+            break
+        output_image_id += 1
+    if update and selected_output_image is not None:
+        st.session_state["input_image"] = selected_output_image
+    return selected_output_image is not None
+
+
+def apply_stroke_to_image(stroke_image, image):
+    image = np.array(image.convert("RGB")).astype(np.float32)
+    height, width, _ = image.shape
+
+    stroke_image = np.array(Image.fromarray(stroke_image).resize((width, height))).astype(np.float32)
+    weight = stroke_image[:, :, -1:] / 255
+    stroke_image = stroke_image[:, :, :-1]
+
+    image = stroke_image * weight + image * (1 - weight)
+    image = np.clip(image, 0, 255).astype(np.uint8)
+    image = Image.fromarray(image)
+    return image
+
+
+@st.cache_data
+def image2bits(image):
+    image_byte = io.BytesIO()
+    image.save(image_byte, format="PNG")
+    image_byte = image_byte.getvalue()
+    return image_byte
+
+
+def show_output_image(image):
+    st.image(image, use_column_width="always")
+    st.button("Use it as input image", key=f"use_output_as_input_{image_id}")
+    st.download_button("Download", data=image2bits(image), file_name="image.png", mime="image/png", key=f"download_output_{image_id}")
+
+
+column_input, column_output = st.columns(2)
+with st.sidebar:
+    # Select a model
+    with st.expander("Model", expanded=True):
+        model_type = st.selectbox("Model type", [model_type_ for model_type_ in config])
+        fixed_parameters = config[model_type]["fixed_parameters"]
+        model_path_list = ["None"] + load_model_list(model_type)
+        model_path = st.selectbox("Model path", model_path_list)
+
+        # Load the model
+        if model_path == "None":
+            # No models are selected. Release VRAM.
+            st.markdown("No models are selected.")
+            release_model()
+        else:
+            # A model is selected.
+            model_path = os.path.join(config[model_type]["model_folder"], model_path)
+            if st.session_state.get("loaded_model_path", "") != model_path:
+                # The loaded model is not the selected model. Reload it.
+                st.markdown(f"Loading model at {model_path}.")
+                st.markdown("Please wait a moment...")
+                release_model()
+                model_manager, pipeline = load_model(model_type, model_path)
+                st.markdown("Done.")
+            else:
+                # The loaded model is not the selected model. Fetch it from `st.session_state`.
+                st.markdown(f"Loading model at {model_path}.")
+                st.markdown("Please wait a moment...")
+                model_manager, pipeline = st.session_state.model_manager, st.session_state.pipeline
+                st.markdown("Done.")
+
+    # Show parameters
+    with st.expander("Prompt", expanded=True):
+        prompt = st.text_area("Positive prompt")
+        if "negative_prompt" in fixed_parameters:
+            negative_prompt = fixed_parameters["negative_prompt"]
+        else:
+            negative_prompt = st.text_area("Negative prompt")
+        if "cfg_scale" in fixed_parameters:
+            cfg_scale = fixed_parameters["cfg_scale"]
+        else:
+            cfg_scale = st.slider("Classifier-free guidance scale", min_value=1.0, max_value=10.0, value=7.5)
+    with st.expander("Image", expanded=True):
+        if "num_inference_steps" in fixed_parameters:
+            num_inference_steps = fixed_parameters["num_inference_steps"]
+        else:
+            num_inference_steps = st.slider("Inference steps", min_value=1, max_value=100, value=20)
+        if "height" in fixed_parameters:
+            height = fixed_parameters["height"]
+        else:
+            height = st.select_slider("Height", options=[256, 512, 768, 1024, 2048], value=512)
+        if "width" in fixed_parameters:
+            width = fixed_parameters["width"]
+        else:
+            width = st.select_slider("Width", options=[256, 512, 768, 1024, 2048], value=512)
+        num_images = st.number_input("Number of images", value=2)
+        use_fixed_seed = st.checkbox("Use fixed seed", value=False)
+        if use_fixed_seed:
+            seed = st.number_input("Random seed", min_value=0, max_value=10**9, step=1, value=0)
+
+    # Other fixed parameters
+    denoising_strength = 1.0
+    repetition = 1
+
+
+# Show input image
+with column_input:
+    with st.expander("Input image (Optional)", expanded=True):
+        with st.container(border=True):
+            column_white_board, column_upload_image = st.columns([1, 2])
+            with column_white_board:
+                create_white_board = st.button("Create white board")
+                delete_input_image = st.button("Delete input image")
+            with column_upload_image:
+                upload_image = st.file_uploader("Upload image", type=["png", "jpg"], key="upload_image")
+
+        if upload_image is not None:
+            st.session_state["input_image"] = crop_and_resize(Image.open(upload_image), height, width)
+        elif create_white_board:
+            st.session_state["input_image"] = Image.fromarray(np.ones((height, width, 3), dtype=np.uint8) * 255)
+        else:
+            use_output_image_as_input()
+
+        if delete_input_image and "input_image" in st.session_state:
+            del st.session_state.input_image
+        if delete_input_image and "upload_image" in st.session_state:
+            del st.session_state.upload_image
+
+        input_image = st.session_state.get("input_image", None)
+        if input_image is not None:
+            with st.container(border=True):
+                column_drawing_mode, column_color_1, column_color_2 = st.columns([4, 1, 1])
+                with column_drawing_mode:
+                    drawing_mode = st.radio("Drawing tool", ["transform", "freedraw", "line", "rect"], horizontal=True, index=1)
+                with column_color_1:
+                    stroke_color = st.color_picker("Stroke color")
+                with column_color_2:
+                    fill_color = st.color_picker("Fill color")
+                stroke_width = st.slider("Stroke width", min_value=1, max_value=50, value=10)
+            with st.container(border=True):
+                denoising_strength = st.slider("Denoising strength", min_value=0.0, max_value=1.0, value=0.7)
+                repetition = st.slider("Repetition", min_value=1, max_value=8, value=1)
+            with st.container(border=True):
+                input_width, input_height = input_image.size
+                canvas_result = st_canvas(
+                    fill_color=fill_color,
+                    stroke_width=stroke_width,
+                    stroke_color=stroke_color,
+                    background_color="rgba(255, 255, 255, 0)",
+                    background_image=input_image,
+                    update_streamlit=True,
+                    height=int(512 / input_width * input_height),
+                    width=512,
+                    drawing_mode=drawing_mode,
+                    key="canvas"
+                )
+
+    num_painter_layer = st.number_input("Number of painter layers", min_value=0, max_value=10, step=1, value=0)
+    local_prompts, masks, mask_scales = [], [], []
+    white_board = Image.fromarray(np.ones((512, 512, 3), dtype=np.uint8) * 255)
+    painter_layers_json_data = []
+    for painter_tab_id in range(num_painter_layer):
+        with st.expander(f"Painter layer {painter_tab_id}", expanded=True):
+            enable_local_prompt = st.checkbox(f"Enable prompt {painter_tab_id}", value=True)
+            local_prompt = st.text_area(f"Prompt {painter_tab_id}")
+            mask_scale = st.slider(f"Mask scale {painter_tab_id}", min_value=0.0, max_value=3.0, value=1.0)
+            stroke_width = st.slider(f"Stroke width {painter_tab_id}", min_value=1, max_value=300, value=100)
+            canvas_result_local = st_canvas(
+                fill_color="#000000",
+                stroke_width=stroke_width,
+                stroke_color="#000000",
+                background_color="rgba(255, 255, 255, 0)",
+                background_image=white_board,
+                update_streamlit=True,
+                height=512,
+                width=512,
+                drawing_mode="freedraw",
+                key=f"canvas_{painter_tab_id}"
+            )
+            if canvas_result_local.json_data is not None:
+                painter_layers_json_data.append(canvas_result_local.json_data.copy())
+                painter_layers_json_data[-1]["prompt"] = local_prompt
+            if enable_local_prompt:
+                local_prompts.append(local_prompt)
+                if canvas_result_local.image_data is not None:
+                    mask = apply_stroke_to_image(canvas_result_local.image_data, white_board)
+                else:
+                    mask = white_board
+                mask = Image.fromarray(255 - np.array(mask))
+                masks.append(mask)
+                mask_scales.append(mask_scale)
+    save_painter_layers = st.button("Save painter layers")
+    if save_painter_layers:
+        os.makedirs("data/painter_layers", exist_ok=True)
+        json_file_path = f"data/painter_layers/{time.time_ns()}.json"
+        with open(json_file_path, "w") as f:
+            json.dump(painter_layers_json_data, f, indent=4)
+            st.markdown(f"Painter layers are saved in {json_file_path}.")
+
+
+with column_output:
+    run_button = st.button("Generate image", type="primary")
+    auto_update = st.checkbox("Auto update", value=False)
+    num_image_columns = st.slider("Columns", min_value=1, max_value=8, value=2)
+    image_columns = st.columns(num_image_columns)
+
+    # Run
+    if (run_button or auto_update) and model_path != "None":
+
+        if input_image is not None:
+            input_image = input_image.resize((width, height))
+            if canvas_result.image_data is not None:
+                input_image = apply_stroke_to_image(canvas_result.image_data, input_image)
+
+        output_images = []
+        for image_id in range(num_images * repetition):
+            if use_fixed_seed:
+                torch.manual_seed(seed + image_id)
+            else:
+                torch.manual_seed(np.random.randint(0, 10**9))
+            if image_id >= num_images:
+                input_image = output_images[image_id - num_images]
+            with image_columns[image_id % num_image_columns]:
+                progress_bar_st = st.progress(0.0)
+                image = pipeline(
+                    prompt, negative_prompt=negative_prompt,
+                    local_prompts=local_prompts, masks=masks, mask_scales=mask_scales,
+                    cfg_scale=cfg_scale, num_inference_steps=num_inference_steps,
+                    height=height, width=width,
+                    input_image=input_image, denoising_strength=denoising_strength,
+                    progress_bar_st=progress_bar_st
+                )
+                output_images.append(image)
+                progress_bar_st.progress(1.0)
+                show_output_image(image)
+                st.session_state["output_images"] = output_images
+
+    elif "output_images" in st.session_state:
+        for image_id in range(len(st.session_state.output_images)):
+            with image_columns[image_id % num_image_columns]:
+                image = st.session_state.output_images[image_id]
+                progress_bar = st.progress(1.0)
+                show_output_image(image)
+    if "upload_image" in st.session_state and use_output_image_as_input(update=False):
+        st.markdown("If you want to use an output image as input image, please delete the uploaded image manually.")

apps/streamlit/pages/2_Video_Creator.py

@@ -0,0 +1,197 @@
+import streamlit as st
+st.set_page_config(layout="wide")
+from diffsynth import SDVideoPipelineRunner
+import os
+import numpy as np
+
+
+def load_model_list(folder):
+    file_list = os.listdir(folder)
+    file_list = [i for i in file_list if i.endswith(".safetensors") or i.endswith(".pth") or i.endswith(".ckpt")]
+    file_list = sorted(file_list)
+    return file_list
+
+
+def match_processor_id(model_name, supported_processor_id_list):
+    sorted_processor_id = [i[1] for i in sorted([(-len(i), i) for i in supported_processor_id_list])]
+    for processor_id in sorted_processor_id:
+        if processor_id in model_name:
+            return supported_processor_id_list.index(processor_id) + 1
+    return 0
+
+
+config = {
+    "models": {
+        "model_list": [],
+        "textual_inversion_folder": "models/textual_inversion",
+        "device": "cuda",
+        "lora_alphas": [],
+        "controlnet_units": []
+    },
+    "data": {
+        "input_frames": None,
+        "controlnet_frames": [],
+        "output_folder": "output",
+        "fps": 60
+    },
+    "pipeline": {
+        "seed": 0,
+        "pipeline_inputs": {}
+    }
+}
+
+
+with st.expander("Model", expanded=True):
+    stable_diffusion_ckpt = st.selectbox("Stable Diffusion", ["None"] + load_model_list("models/stable_diffusion"))
+    if stable_diffusion_ckpt != "None":
+        config["models"]["model_list"].append(os.path.join("models/stable_diffusion", stable_diffusion_ckpt))
+    animatediff_ckpt = st.selectbox("AnimateDiff", ["None"] + load_model_list("models/AnimateDiff"))
+    if animatediff_ckpt != "None":
+        config["models"]["model_list"].append(os.path.join("models/AnimateDiff", animatediff_ckpt))
+    column_lora, column_lora_alpha = st.columns([2, 1])
+    with column_lora:
+        sd_lora_ckpt = st.selectbox("LoRA", ["None"] + load_model_list("models/lora"))
+    with column_lora_alpha:
+        lora_alpha = st.slider("LoRA Alpha", min_value=-4.0, max_value=4.0, value=1.0, step=0.1)
+    if sd_lora_ckpt != "None":
+        config["models"]["model_list"].append(os.path.join("models/lora", sd_lora_ckpt))
+        config["models"]["lora_alphas"].append(lora_alpha)
+
+
+with st.expander("Data", expanded=True):
+    with st.container(border=True):
+        input_video = st.text_input("Input Video File Path (e.g., data/your_video.mp4)", value="")
+        column_height, column_width, column_start_frame_index, column_end_frame_index = st.columns([2, 2, 1, 1])
+        with column_height:
+            height = st.select_slider("Height", options=[256, 512, 768, 1024, 1536, 2048], value=1024)
+        with column_width:
+            width = st.select_slider("Width", options=[256, 512, 768, 1024, 1536, 2048], value=1024)
+        with column_start_frame_index:
+            start_frame_id = st.number_input("Start Frame id", value=0)
+        with column_end_frame_index:
+            end_frame_id = st.number_input("End Frame id", value=16)
+        if input_video != "":
+            config["data"]["input_frames"] = {
+                "video_file": input_video,
+                "image_folder": None,
+                "height": height,
+                "width": width,
+                "start_frame_id": start_frame_id,
+                "end_frame_id": end_frame_id
+            }
+    with st.container(border=True):
+        output_video = st.text_input("Output Video File Path (e.g., data/a_folder_to_save_something)", value="output")
+        fps = st.number_input("FPS", value=60)
+        config["data"]["output_folder"] = output_video
+        config["data"]["fps"] = fps
+
+
+with st.expander("ControlNet Units", expanded=True):
+    supported_processor_id_list = ["canny", "depth", "softedge", "lineart", "lineart_anime", "openpose", "tile"]
+    controlnet_units = st.tabs(["ControlNet Unit 0", "ControlNet Unit 1", "ControlNet Unit 2"])
+    for controlnet_id in range(len(controlnet_units)):
+        with controlnet_units[controlnet_id]:
+            controlnet_ckpt = st.selectbox("ControlNet", ["None"] + load_model_list("models/ControlNet"),
+                                        key=f"controlnet_ckpt_{controlnet_id}")
+            processor_id = st.selectbox("Processor", ["None"] + supported_processor_id_list,
+                                        index=match_processor_id(controlnet_ckpt, supported_processor_id_list),
+                                        disabled=controlnet_ckpt == "None", key=f"processor_id_{controlnet_id}")
+            controlnet_scale = st.slider("Scale", min_value=0.0, max_value=1.0, step=0.01, value=0.5,
+                                        disabled=controlnet_ckpt == "None", key=f"controlnet_scale_{controlnet_id}")
+            use_input_video_as_controlnet_input = st.checkbox("Use input video as ControlNet input", value=True,
+                                                              disabled=controlnet_ckpt == "None",
+                                                              key=f"use_input_video_as_controlnet_input_{controlnet_id}")
+            if not use_input_video_as_controlnet_input:
+                controlnet_input_video = st.text_input("ControlNet Input Video File Path", value="",
+                                            disabled=controlnet_ckpt == "None", key=f"controlnet_input_video_{controlnet_id}")
+                column_height, column_width, column_start_frame_index, column_end_frame_index = st.columns([2, 2, 1, 1])
+                with column_height:
+                    height = st.select_slider("Height", options=[256, 512, 768, 1024, 1536, 2048], value=1024,
+                                              disabled=controlnet_ckpt == "None", key=f"controlnet_height_{controlnet_id}")
+                with column_width:
+                    width = st.select_slider("Width", options=[256, 512, 768, 1024, 1536, 2048], value=1024,
+                                              disabled=controlnet_ckpt == "None", key=f"controlnet_width_{controlnet_id}")
+                with column_start_frame_index:
+                    start_frame_id = st.number_input("Start Frame id", value=0,
+                                                     disabled=controlnet_ckpt == "None", key=f"controlnet_start_frame_id_{controlnet_id}")
+                with column_end_frame_index:
+                    end_frame_id = st.number_input("End Frame id", value=16,
+                                                   disabled=controlnet_ckpt == "None", key=f"controlnet_end_frame_id_{controlnet_id}")
+                if input_video != "":
+                    config["data"]["input_video"] = {
+                        "video_file": input_video,
+                        "image_folder": None,
+                        "height": height,
+                        "width": width,
+                        "start_frame_id": start_frame_id,
+                        "end_frame_id": end_frame_id
+                    }
+            if controlnet_ckpt != "None":
+                config["models"]["model_list"].append(os.path.join("models/ControlNet", controlnet_ckpt))
+                config["models"]["controlnet_units"].append({
+                    "processor_id": processor_id,
+                    "model_path": os.path.join("models/ControlNet", controlnet_ckpt),
+                    "scale": controlnet_scale,
+                })
+                if use_input_video_as_controlnet_input:
+                    config["data"]["controlnet_frames"].append(config["data"]["input_frames"])
+                else:
+                    config["data"]["controlnet_frames"].append({
+                        "video_file": input_video,
+                        "image_folder": None,
+                        "height": height,
+                        "width": width,
+                        "start_frame_id": start_frame_id,
+                        "end_frame_id": end_frame_id
+                    })
+
+
+with st.container(border=True):
+    with st.expander("Seed", expanded=True):
+        use_fixed_seed = st.checkbox("Use fixed seed", value=False)
+        if use_fixed_seed:
+            seed = st.number_input("Random seed", min_value=0, max_value=10**9, step=1, value=0)
+        else:
+            seed = np.random.randint(0, 10**9)
+    with st.expander("Textual Guidance", expanded=True):
+        prompt = st.text_area("Positive prompt")
+        negative_prompt = st.text_area("Negative prompt")
+        column_cfg_scale, column_clip_skip = st.columns(2)
+        with column_cfg_scale:
+            cfg_scale = st.slider("Classifier-free guidance scale", min_value=1.0, max_value=10.0, value=7.0)
+        with column_clip_skip:
+            clip_skip = st.slider("Clip Skip", min_value=1, max_value=4, value=1)
+    with st.expander("Denoising", expanded=True):
+        column_num_inference_steps, column_denoising_strength = st.columns(2)
+        with column_num_inference_steps:
+            num_inference_steps = st.slider("Inference steps", min_value=1, max_value=100, value=10)
+        with column_denoising_strength:
+            denoising_strength = st.slider("Denoising strength", min_value=0.0, max_value=1.0, value=1.0)
+    with st.expander("Efficiency", expanded=False):
+        animatediff_batch_size = st.slider("Animatediff batch size (sliding window size)", min_value=1, max_value=32, value=16, step=1)
+        animatediff_stride = st.slider("Animatediff stride",
+                                       min_value=1,
+                                       max_value=max(2, animatediff_batch_size),
+                                       value=max(1, animatediff_batch_size // 2),
+                                       step=1)
+        unet_batch_size = st.slider("UNet batch size", min_value=1, max_value=32, value=1, step=1)
+        controlnet_batch_size = st.slider("ControlNet batch size", min_value=1, max_value=32, value=1, step=1)
+        cross_frame_attention = st.checkbox("Enable Cross-Frame Attention", value=False)
+    config["pipeline"]["seed"] = seed
+    config["pipeline"]["pipeline_inputs"] = {
+        "prompt": prompt,
+        "negative_prompt": negative_prompt,
+        "cfg_scale": cfg_scale,
+        "clip_skip": clip_skip,
+        "denoising_strength": denoising_strength,
+        "num_inference_steps": num_inference_steps,
+        "animatediff_batch_size": animatediff_batch_size,
+        "animatediff_stride": animatediff_stride,
+        "unet_batch_size": unet_batch_size,
+        "controlnet_batch_size": controlnet_batch_size,
+        "cross_frame_attention": cross_frame_attention,
+    }
+
+run_button = st.button("☢️Run☢️", type="primary")
+if run_button:
+    SDVideoPipelineRunner(in_streamlit=True).run(config)

diffsynth.egg-info/PKG-INFO

@@ -0,0 +1,23 @@
+Metadata-Version: 2.1
+Name: diffsynth
+Version: 1.1.2
+Summary: Enjoy the magic of Diffusion models!
+Author: Artiprocher
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Operating System :: OS Independent
+Requires-Python: >=3.6
+License-File: LICENSE
+Requires-Dist: torch>=2.0.0
+Requires-Dist: torchvision
+Requires-Dist: cupy-cuda12x
+Requires-Dist: transformers==4.46.2
+Requires-Dist: controlnet-aux==0.0.7
+Requires-Dist: imageio
+Requires-Dist: imageio[ffmpeg]
+Requires-Dist: safetensors
+Requires-Dist: einops
+Requires-Dist: sentencepiece
+Requires-Dist: protobuf
+Requires-Dist: modelscope
+Requires-Dist: ftfy

diffsynth.egg-info/SOURCES.txt

@@ -0,0 +1,215 @@
+LICENSE
+README.md
+setup.py
+diffsynth/__init__.py
+diffsynth.egg-info/PKG-INFO
+diffsynth.egg-info/SOURCES.txt
+diffsynth.egg-info/dependency_links.txt
+diffsynth.egg-info/requires.txt
+diffsynth.egg-info/top_level.txt
+diffsynth/configs/__init__.py
+diffsynth/configs/model_config.py
+diffsynth/controlnets/__init__.py
+diffsynth/controlnets/controlnet_unit.py
+diffsynth/controlnets/processors.py
+diffsynth/data/__init__.py
+diffsynth/data/simple_text_image.py
+diffsynth/data/video.py
+diffsynth/extensions/__init__.py
+diffsynth/extensions/ESRGAN/__init__.py
+diffsynth/extensions/FastBlend/__init__.py
+diffsynth/extensions/FastBlend/api.py
+diffsynth/extensions/FastBlend/cupy_kernels.py
+diffsynth/extensions/FastBlend/data.py
+diffsynth/extensions/FastBlend/patch_match.py
+diffsynth/extensions/FastBlend/runners/__init__.py
+diffsynth/extensions/FastBlend/runners/accurate.py
+diffsynth/extensions/FastBlend/runners/balanced.py
+diffsynth/extensions/FastBlend/runners/fast.py
+diffsynth/extensions/FastBlend/runners/interpolation.py
+diffsynth/extensions/ImageQualityMetric/__init__.py
+diffsynth/extensions/ImageQualityMetric/aesthetic.py
+diffsynth/extensions/ImageQualityMetric/clip.py
+diffsynth/extensions/ImageQualityMetric/config.py
+diffsynth/extensions/ImageQualityMetric/hps.py
+diffsynth/extensions/ImageQualityMetric/imagereward.py
+diffsynth/extensions/ImageQualityMetric/mps.py
+diffsynth/extensions/ImageQualityMetric/pickscore.py
+diffsynth/extensions/ImageQualityMetric/BLIP/__init__.py
+diffsynth/extensions/ImageQualityMetric/BLIP/blip.py
+diffsynth/extensions/ImageQualityMetric/BLIP/blip_pretrain.py
+diffsynth/extensions/ImageQualityMetric/BLIP/med.py
+diffsynth/extensions/ImageQualityMetric/BLIP/vit.py
+diffsynth/extensions/ImageQualityMetric/open_clip/__init__.py
+diffsynth/extensions/ImageQualityMetric/open_clip/coca_model.py
+diffsynth/extensions/ImageQualityMetric/open_clip/constants.py
+diffsynth/extensions/ImageQualityMetric/open_clip/factory.py
+diffsynth/extensions/ImageQualityMetric/open_clip/generation_utils.py
+diffsynth/extensions/ImageQualityMetric/open_clip/hf_configs.py
+diffsynth/extensions/ImageQualityMetric/open_clip/hf_model.py
+diffsynth/extensions/ImageQualityMetric/open_clip/loss.py
+diffsynth/extensions/ImageQualityMetric/open_clip/model.py
+diffsynth/extensions/ImageQualityMetric/open_clip/modified_resnet.py
+diffsynth/extensions/ImageQualityMetric/open_clip/openai.py
+diffsynth/extensions/ImageQualityMetric/open_clip/pretrained.py
+diffsynth/extensions/ImageQualityMetric/open_clip/push_to_hf_hub.py
+diffsynth/extensions/ImageQualityMetric/open_clip/timm_model.py
+diffsynth/extensions/ImageQualityMetric/open_clip/tokenizer.py
+diffsynth/extensions/ImageQualityMetric/open_clip/transform.py
+diffsynth/extensions/ImageQualityMetric/open_clip/transformer.py
+diffsynth/extensions/ImageQualityMetric/open_clip/utils.py
+diffsynth/extensions/ImageQualityMetric/open_clip/version.py
+diffsynth/extensions/ImageQualityMetric/trainer/__init__.py
+diffsynth/extensions/ImageQualityMetric/trainer/models/__init__.py
+diffsynth/extensions/ImageQualityMetric/trainer/models/base_model.py
+diffsynth/extensions/ImageQualityMetric/trainer/models/clip_model.py
+diffsynth/extensions/ImageQualityMetric/trainer/models/cross_modeling.py
+diffsynth/extensions/RIFE/__init__.py
+diffsynth/models/__init__.py
+diffsynth/models/attention.py
+diffsynth/models/cog_dit.py
+diffsynth/models/cog_vae.py
+diffsynth/models/downloader.py
+diffsynth/models/flux_controlnet.py
+diffsynth/models/flux_dit.py
+diffsynth/models/flux_ipadapter.py
+diffsynth/models/flux_text_encoder.py
+diffsynth/models/flux_vae.py
+diffsynth/models/hunyuan_dit.py
+diffsynth/models/hunyuan_dit_text_encoder.py
+diffsynth/models/hunyuan_video_dit.py
+diffsynth/models/hunyuan_video_text_encoder.py
+diffsynth/models/hunyuan_video_vae_decoder.py
+diffsynth/models/hunyuan_video_vae_encoder.py
+diffsynth/models/kolors_text_encoder.py
+diffsynth/models/lora.py
+diffsynth/models/model_manager.py
+diffsynth/models/omnigen.py
+diffsynth/models/sd3_dit.py
+diffsynth/models/sd3_text_encoder.py
+diffsynth/models/sd3_vae_decoder.py
+diffsynth/models/sd3_vae_encoder.py
+diffsynth/models/sd_controlnet.py
+diffsynth/models/sd_ipadapter.py
+diffsynth/models/sd_motion.py
+diffsynth/models/sd_text_encoder.py
+diffsynth/models/sd_unet.py
+diffsynth/models/sd_vae_decoder.py
+diffsynth/models/sd_vae_encoder.py
+diffsynth/models/sdxl_controlnet.py
+diffsynth/models/sdxl_ipadapter.py
+diffsynth/models/sdxl_motion.py
+diffsynth/models/sdxl_text_encoder.py
+diffsynth/models/sdxl_unet.py
+diffsynth/models/sdxl_vae_decoder.py
+diffsynth/models/sdxl_vae_encoder.py
+diffsynth/models/stepvideo_dit.py
+diffsynth/models/stepvideo_text_encoder.py
+diffsynth/models/stepvideo_vae.py
+diffsynth/models/svd_image_encoder.py
+diffsynth/models/svd_unet.py
+diffsynth/models/svd_vae_decoder.py
+diffsynth/models/svd_vae_encoder.py
+diffsynth/models/tiler.py
+diffsynth/models/utils.py
+diffsynth/models/wan_video_dit.py
+diffsynth/models/wan_video_image_encoder.py
+diffsynth/models/wan_video_text_encoder.py
+diffsynth/models/wan_video_vae.py
+diffsynth/pipelines/__init__.py
+diffsynth/pipelines/base.py
+diffsynth/pipelines/cog_video.py
+diffsynth/pipelines/dancer.py
+diffsynth/pipelines/flux_image.py
+diffsynth/pipelines/hunyuan_image.py
+diffsynth/pipelines/hunyuan_video.py
+diffsynth/pipelines/omnigen_image.py
+diffsynth/pipelines/pipeline_runner.py
+diffsynth/pipelines/sd3_image.py
+diffsynth/pipelines/sd_image.py
+diffsynth/pipelines/sd_video.py
+diffsynth/pipelines/sdxl_image.py
+diffsynth/pipelines/sdxl_video.py
+diffsynth/pipelines/step_video.py
+diffsynth/pipelines/svd_video.py
+diffsynth/pipelines/wan_video.py
+diffsynth/processors/FastBlend.py
+diffsynth/processors/PILEditor.py
+diffsynth/processors/RIFE.py
+diffsynth/processors/__init__.py
+diffsynth/processors/base.py
+diffsynth/processors/sequencial_processor.py
+diffsynth/prompters/__init__.py
+diffsynth/prompters/base_prompter.py
+diffsynth/prompters/cog_prompter.py
+diffsynth/prompters/flux_prompter.py
+diffsynth/prompters/hunyuan_dit_prompter.py
+diffsynth/prompters/hunyuan_video_prompter.py
+diffsynth/prompters/kolors_prompter.py
+diffsynth/prompters/omnigen_prompter.py
+diffsynth/prompters/omost.py
+diffsynth/prompters/prompt_refiners.py
+diffsynth/prompters/sd3_prompter.py
+diffsynth/prompters/sd_prompter.py
+diffsynth/prompters/sdxl_prompter.py
+diffsynth/prompters/stepvideo_prompter.py
+diffsynth/prompters/wan_prompter.py
+diffsynth/schedulers/__init__.py
+diffsynth/schedulers/continuous_ode.py
+diffsynth/schedulers/ddim.py
+diffsynth/schedulers/flow_match.py
+diffsynth/tokenizer_configs/__init__.py
+diffsynth/tokenizer_configs/cog/tokenizer/added_tokens.json
+diffsynth/tokenizer_configs/cog/tokenizer/special_tokens_map.json
+diffsynth/tokenizer_configs/cog/tokenizer/spiece.model
+diffsynth/tokenizer_configs/cog/tokenizer/tokenizer_config.json
+diffsynth/tokenizer_configs/flux/tokenizer_1/merges.txt
+diffsynth/tokenizer_configs/flux/tokenizer_1/special_tokens_map.json
+diffsynth/tokenizer_configs/flux/tokenizer_1/tokenizer_config.json
+diffsynth/tokenizer_configs/flux/tokenizer_1/vocab.json
+diffsynth/tokenizer_configs/flux/tokenizer_2/special_tokens_map.json
+diffsynth/tokenizer_configs/flux/tokenizer_2/spiece.model
+diffsynth/tokenizer_configs/flux/tokenizer_2/tokenizer.json
+diffsynth/tokenizer_configs/flux/tokenizer_2/tokenizer_config.json
+diffsynth/tokenizer_configs/hunyuan_dit/tokenizer/special_tokens_map.json
+diffsynth/tokenizer_configs/hunyuan_dit/tokenizer/tokenizer_config.json
+diffsynth/tokenizer_configs/hunyuan_dit/tokenizer/vocab.txt
+diffsynth/tokenizer_configs/hunyuan_dit/tokenizer/vocab_org.txt
+diffsynth/tokenizer_configs/hunyuan_dit/tokenizer_t5/config.json
+diffsynth/tokenizer_configs/hunyuan_dit/tokenizer_t5/special_tokens_map.json
+diffsynth/tokenizer_configs/hunyuan_dit/tokenizer_t5/spiece.model
+diffsynth/tokenizer_configs/hunyuan_dit/tokenizer_t5/tokenizer_config.json
+diffsynth/tokenizer_configs/hunyuan_video/tokenizer_1/merges.txt
+diffsynth/tokenizer_configs/hunyuan_video/tokenizer_1/special_tokens_map.json
+diffsynth/tokenizer_configs/hunyuan_video/tokenizer_1/tokenizer_config.json
+diffsynth/tokenizer_configs/hunyuan_video/tokenizer_1/vocab.json
+diffsynth/tokenizer_configs/hunyuan_video/tokenizer_2/special_tokens_map.json
+diffsynth/tokenizer_configs/hunyuan_video/tokenizer_2/tokenizer.json
+diffsynth/tokenizer_configs/hunyuan_video/tokenizer_2/tokenizer_config.json
+diffsynth/tokenizer_configs/kolors/tokenizer/tokenizer.model
+diffsynth/tokenizer_configs/kolors/tokenizer/tokenizer_config.json
+diffsynth/tokenizer_configs/kolors/tokenizer/vocab.txt
+diffsynth/tokenizer_configs/stable_diffusion/tokenizer/merges.txt
+diffsynth/tokenizer_configs/stable_diffusion/tokenizer/special_tokens_map.json
+diffsynth/tokenizer_configs/stable_diffusion/tokenizer/tokenizer_config.json
+diffsynth/tokenizer_configs/stable_diffusion/tokenizer/vocab.json
+diffsynth/tokenizer_configs/stable_diffusion_3/tokenizer_1/merges.txt
+diffsynth/tokenizer_configs/stable_diffusion_3/tokenizer_1/special_tokens_map.json
+diffsynth/tokenizer_configs/stable_diffusion_3/tokenizer_1/tokenizer_config.json
+diffsynth/tokenizer_configs/stable_diffusion_3/tokenizer_1/vocab.json
+diffsynth/tokenizer_configs/stable_diffusion_3/tokenizer_2/merges.txt
+diffsynth/tokenizer_configs/stable_diffusion_3/tokenizer_2/special_tokens_map.json
+diffsynth/tokenizer_configs/stable_diffusion_3/tokenizer_2/tokenizer_config.json
+diffsynth/tokenizer_configs/stable_diffusion_3/tokenizer_2/vocab.json
+diffsynth/tokenizer_configs/stable_diffusion_3/tokenizer_3/special_tokens_map.json
+diffsynth/tokenizer_configs/stable_diffusion_3/tokenizer_3/spiece.model
+diffsynth/tokenizer_configs/stable_diffusion_3/tokenizer_3/tokenizer.json
+diffsynth/tokenizer_configs/stable_diffusion_3/tokenizer_3/tokenizer_config.json
+diffsynth/tokenizer_configs/stable_diffusion_xl/tokenizer_2/merges.txt
+diffsynth/tokenizer_configs/stable_diffusion_xl/tokenizer_2/special_tokens_map.json
+diffsynth/tokenizer_configs/stable_diffusion_xl/tokenizer_2/tokenizer_config.json
+diffsynth/tokenizer_configs/stable_diffusion_xl/tokenizer_2/vocab.json
+diffsynth/trainers/__init__.py
+diffsynth/trainers/text_to_image.py
+diffsynth/vram_management/__init__.py
+diffsynth/vram_management/layers.py

diffsynth.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

diffsynth.egg-info/requires.txt

@@ -0,0 +1,13 @@
+torch>=2.0.0
+torchvision
+cupy-cuda12x
+transformers==4.46.2
+controlnet-aux==0.0.7
+imageio
+imageio[ffmpeg]
+safetensors
+einops
+sentencepiece
+protobuf
+modelscope
+ftfy

diffsynth.egg-info/top_level.txt

@@ -0,0 +1 @@
+diffsynth

diffsynth/__init__.py

@@ -0,0 +1,6 @@
+from .data import *
+from .models import *
+from .prompters import *
+from .schedulers import *
+from .pipelines import *
+from .controlnets import *

diffsynth/configs/model_config.py

@@ -0,0 +1,756 @@
+from typing_extensions import Literal, TypeAlias
+
+from ..models.sd_text_encoder import SDTextEncoder
+from ..models.sd_unet import SDUNet
+from ..models.sd_vae_encoder import SDVAEEncoder
+from ..models.sd_vae_decoder import SDVAEDecoder
+
+from ..models.sdxl_text_encoder import SDXLTextEncoder, SDXLTextEncoder2
+from ..models.sdxl_unet import SDXLUNet
+from ..models.sdxl_vae_decoder import SDXLVAEDecoder
+from ..models.sdxl_vae_encoder import SDXLVAEEncoder
+
+from ..models.sd3_text_encoder import SD3TextEncoder1, SD3TextEncoder2, SD3TextEncoder3
+from ..models.sd3_dit import SD3DiT
+from ..models.sd3_vae_decoder import SD3VAEDecoder
+from ..models.sd3_vae_encoder import SD3VAEEncoder
+
+from ..models.sd_controlnet import SDControlNet
+from ..models.sdxl_controlnet import SDXLControlNetUnion
+
+from ..models.sd_motion import SDMotionModel
+from ..models.sdxl_motion import SDXLMotionModel
+
+from ..models.svd_image_encoder import SVDImageEncoder
+from ..models.svd_unet import SVDUNet
+from ..models.svd_vae_decoder import SVDVAEDecoder
+from ..models.svd_vae_encoder import SVDVAEEncoder
+
+from ..models.sd_ipadapter import SDIpAdapter, IpAdapterCLIPImageEmbedder
+from ..models.sdxl_ipadapter import SDXLIpAdapter, IpAdapterXLCLIPImageEmbedder
+
+from ..models.hunyuan_dit_text_encoder import HunyuanDiTCLIPTextEncoder, HunyuanDiTT5TextEncoder
+from ..models.hunyuan_dit import HunyuanDiT
+
+from ..models.flux_dit import FluxDiT
+from ..models.flux_text_encoder import FluxTextEncoder2
+from ..models.flux_vae import FluxVAEEncoder, FluxVAEDecoder
+from ..models.flux_controlnet import FluxControlNet
+from ..models.flux_ipadapter import FluxIpAdapter
+
+from ..models.cog_vae import CogVAEEncoder, CogVAEDecoder
+from ..models.cog_dit import CogDiT
+
+from ..models.omnigen import OmniGenTransformer
+
+from ..models.hunyuan_video_vae_decoder import HunyuanVideoVAEDecoder
+from ..models.hunyuan_video_vae_encoder import HunyuanVideoVAEEncoder
+
+from ..extensions.RIFE import IFNet
+from ..extensions.ESRGAN import RRDBNet
+
+from ..models.hunyuan_video_dit import HunyuanVideoDiT
+
+from ..models.stepvideo_vae import StepVideoVAE
+from ..models.stepvideo_dit import StepVideoModel
+
+from ..models.wan_video_dit import WanModel
+from ..models.wan_video_text_encoder import WanTextEncoder
+from ..models.wan_video_image_encoder import WanImageEncoder
+from ..models.wan_video_vae import WanVideoVAE
+
+
+model_loader_configs = [
+    # These configs are provided for detecting model type automatically.
+    # The format is (state_dict_keys_hash, state_dict_keys_hash_with_shape, model_names, model_classes, model_resource)
+    (None, "091b0e30e77c76626b3ba62acdf95343", ["sd_controlnet"], [SDControlNet], "civitai"),
+    (None, "4a6c8306a27d916dea81263c8c88f450", ["hunyuan_dit_clip_text_encoder"], [HunyuanDiTCLIPTextEncoder], "civitai"),
+    (None, "f4aec400fe394297961218c768004521", ["hunyuan_dit"], [HunyuanDiT], "civitai"),
+    (None, "9e6e58043a5a2e332803ed42f6ee7181", ["hunyuan_dit_t5_text_encoder"], [HunyuanDiTT5TextEncoder], "civitai"),
+    (None, "13115dd45a6e1c39860f91ab073b8a78", ["sdxl_vae_encoder", "sdxl_vae_decoder"], [SDXLVAEEncoder, SDXLVAEDecoder], "diffusers"),
+    (None, "d78aa6797382a6d455362358a3295ea9", ["sd_ipadapter_clip_image_encoder"], [IpAdapterCLIPImageEmbedder], "diffusers"),
+    (None, "e291636cc15e803186b47404262ef812", ["sd_ipadapter"], [SDIpAdapter], "civitai"),
+    (None, "399c81f2f8de8d1843d0127a00f3c224", ["sdxl_ipadapter_clip_image_encoder"], [IpAdapterXLCLIPImageEmbedder], "diffusers"),
+    (None, "a64eac9aa0db4b9602213bc0131281c7", ["sdxl_ipadapter"], [SDXLIpAdapter], "civitai"),
+    (None, "52817e4fdd89df154f02749ca6f692ac", ["sdxl_unet"], [SDXLUNet], "diffusers"),
+    (None, "03343c606f16d834d6411d0902b53636", ["sd_text_encoder", "sd_unet", "sd_vae_decoder", "sd_vae_encoder"], [SDTextEncoder, SDUNet, SDVAEDecoder, SDVAEEncoder], "civitai"),
+    (None, "d4ba77a7ece070679b4a987f58f201e9", ["sd_text_encoder"], [SDTextEncoder], "civitai"),
+    (None, "d0c89e55c5a57cf3981def0cb1c9e65a", ["sd_vae_decoder", "sd_vae_encoder"], [SDVAEDecoder, SDVAEEncoder], "civitai"),
+    (None, "3926bf373b39a67eeafd7901478a47a7", ["sd_unet"], [SDUNet], "civitai"),
+    (None, "1e0c39ec176b9007c05f76d52b554a4d", ["sd3_text_encoder_1", "sd3_text_encoder_2", "sd3_dit", "sd3_vae_encoder", "sd3_vae_decoder"], [SD3TextEncoder1, SD3TextEncoder2, SD3DiT, SD3VAEEncoder, SD3VAEDecoder], "civitai"),
+    (None, "d9e0290829ba8d98e28e1a2b1407db4a", ["sd3_text_encoder_1", "sd3_text_encoder_2", "sd3_text_encoder_3", "sd3_dit", "sd3_vae_encoder", "sd3_vae_decoder"], [SD3TextEncoder1, SD3TextEncoder2, SD3TextEncoder3, SD3DiT, SD3VAEEncoder, SD3VAEDecoder], "civitai"),
+    (None, "5072d0b24e406b49507abe861cf97691", ["sd3_text_encoder_3"], [SD3TextEncoder3], "civitai"),
+    (None, "4cf64a799d04260df438c6f33c9a047e", ["sdxl_text_encoder", "sdxl_text_encoder_2", "sdxl_unet", "sdxl_vae_decoder", "sdxl_vae_encoder"], [SDXLTextEncoder, SDXLTextEncoder2, SDXLUNet, SDXLVAEDecoder, SDXLVAEEncoder], "civitai"),
+    (None, "d9b008a867c498ab12ad24042eff8e3f", ["sdxl_text_encoder", "sdxl_text_encoder_2", "sdxl_unet", "sdxl_vae_decoder", "sdxl_vae_encoder"], [SDXLTextEncoder, SDXLTextEncoder2, SDXLUNet, SDXLVAEDecoder, SDXLVAEEncoder], "civitai"), # SDXL-Turbo
+    (None, "025bb7452e531a3853d951d77c63f032", ["sdxl_text_encoder", "sdxl_text_encoder_2"], [SDXLTextEncoder, SDXLTextEncoder2], "civitai"),
+    (None, "298997b403a4245c04102c9f36aac348", ["sdxl_unet"], [SDXLUNet], "civitai"),
+    (None, "2a07abce74b4bdc696b76254ab474da6", ["svd_image_encoder", "svd_unet", "svd_vae_decoder", "svd_vae_encoder"], [SVDImageEncoder, SVDUNet, SVDVAEDecoder, SVDVAEEncoder], "civitai"),
+    (None, "c96a285a6888465f87de22a984d049fb", ["sd_motion_modules"], [SDMotionModel], "civitai"),
+    (None, "72907b92caed19bdb2adb89aa4063fe2", ["sdxl_motion_modules"], [SDXLMotionModel], "civitai"),
+    (None, "31d2d9614fba60511fc9bf2604aa01f7", ["sdxl_controlnet"], [SDXLControlNetUnion], "diffusers"),
+    (None, "94eefa3dac9cec93cb1ebaf1747d7b78", ["sd3_text_encoder_1"], [SD3TextEncoder1], "diffusers"),
+    (None, "1aafa3cc91716fb6b300cc1cd51b85a3", ["flux_vae_encoder", "flux_vae_decoder"], [FluxVAEEncoder, FluxVAEDecoder], "diffusers"),
+    (None, "21ea55f476dfc4fd135587abb59dfe5d", ["flux_vae_encoder", "flux_vae_decoder"], [FluxVAEEncoder, FluxVAEDecoder], "civitai"),
+    (None, "a29710fea6dddb0314663ee823598e50", ["flux_dit"], [FluxDiT], "civitai"),
+    (None, "57b02550baab820169365b3ee3afa2c9", ["flux_dit"], [FluxDiT], "civitai"),
+    (None, "3394f306c4cbf04334b712bf5aaed95f", ["flux_dit"], [FluxDiT], "civitai"),
+    (None, "023f054d918a84ccf503481fd1e3379e", ["flux_dit"], [FluxDiT], "civitai"),
+    (None, "605c56eab23e9e2af863ad8f0813a25d", ["flux_dit"], [FluxDiT], "diffusers"),
+    (None, "280189ee084bca10f70907bf6ce1649d", ["cog_vae_encoder", "cog_vae_decoder"], [CogVAEEncoder, CogVAEDecoder], "diffusers"),
+    (None, "9b9313d104ac4df27991352fec013fd4", ["rife"], [IFNet], "civitai"),
+    (None, "6b7116078c4170bfbeaedc8fe71f6649", ["esrgan"], [RRDBNet], "civitai"),
+    (None, "61cbcbc7ac11f169c5949223efa960d1", ["omnigen_transformer"], [OmniGenTransformer], "diffusers"),
+    (None, "78d18b9101345ff695f312e7e62538c0", ["flux_controlnet"], [FluxControlNet], "diffusers"),
+    (None, "b001c89139b5f053c715fe772362dd2a", ["flux_controlnet"], [FluxControlNet], "diffusers"),
+    (None, "52357cb26250681367488a8954c271e8", ["flux_controlnet"], [FluxControlNet], "diffusers"),
+    (None, "0cfd1740758423a2a854d67c136d1e8c", ["flux_controlnet"], [FluxControlNet], "diffusers"),
+    (None, "4daaa66cc656a8fe369908693dad0a35", ["flux_ipadapter"], [FluxIpAdapter], "diffusers"),
+    (None, "51aed3d27d482fceb5e0739b03060e8f", ["sd3_dit", "sd3_vae_encoder", "sd3_vae_decoder"], [SD3DiT, SD3VAEEncoder, SD3VAEDecoder], "civitai"),
+    (None, "98cc34ccc5b54ae0e56bdea8688dcd5a", ["sd3_text_encoder_2"], [SD3TextEncoder2], "civitai"),
+    (None, "77ff18050dbc23f50382e45d51a779fe", ["sd3_dit", "sd3_vae_encoder", "sd3_vae_decoder"], [SD3DiT, SD3VAEEncoder, SD3VAEDecoder], "civitai"),
+    (None, "5da81baee73198a7c19e6d2fe8b5148e", ["sd3_text_encoder_1"], [SD3TextEncoder1], "diffusers"),
+    (None, "aeb82dce778a03dcb4d726cb03f3c43f", ["hunyuan_video_vae_decoder", "hunyuan_video_vae_encoder"], [HunyuanVideoVAEDecoder, HunyuanVideoVAEEncoder], "diffusers"),
+    (None, "b9588f02e78f5ccafc9d7c0294e46308", ["hunyuan_video_dit"], [HunyuanVideoDiT], "civitai"),
+    (None, "84ef4bd4757f60e906b54aa6a7815dc6", ["hunyuan_video_dit"], [HunyuanVideoDiT], "civitai"),
+    (None, "68beaf8429b7c11aa8ca05b1bd0058bd", ["stepvideo_vae"], [StepVideoVAE], "civitai"),
+    (None, "5c0216a2132b082c10cb7a0e0377e681", ["stepvideo_dit"], [StepVideoModel], "civitai"),
+    (None, "9269f8db9040a9d860eaca435be61814", ["wan_video_dit"], [WanModel], "civitai"),
+    (None, "aafcfd9672c3a2456dc46e1cb6e52c70", ["wan_video_dit"], [WanModel], "civitai"),
+    (None, "6bfcfb3b342cb286ce886889d519a77e", ["wan_video_dit"], [WanModel], "civitai"),
+    (None, "cb104773c6c2cb6df4f9529ad5c60d0b", ["wan_video_dit"], [WanModel], "diffusers"),
+    (None, "9c8818c2cbea55eca56c7b447df170da", ["wan_video_text_encoder"], [WanTextEncoder], "civitai"),
+    (None, "5941c53e207d62f20f9025686193c40b", ["wan_video_image_encoder"], [WanImageEncoder], "civitai"),
+    (None, "1378ea763357eea97acdef78e65d6d96", ["wan_video_vae"], [WanVideoVAE], "civitai"),
+    (None, "ccc42284ea13e1ad04693284c7a09be6", ["wan_video_vae"], [WanVideoVAE], "civitai"),
+]
+huggingface_model_loader_configs = [
+    # These configs are provided for detecting model type automatically.
+    # The format is (architecture_in_huggingface_config, huggingface_lib, model_name, redirected_architecture)
+    ("ChatGLMModel", "diffsynth.models.kolors_text_encoder", "kolors_text_encoder", None),
+    ("MarianMTModel", "transformers.models.marian.modeling_marian", "translator", None),
+    ("BloomForCausalLM", "transformers.models.bloom.modeling_bloom", "beautiful_prompt", None),
+    ("Qwen2ForCausalLM", "transformers.models.qwen2.modeling_qwen2", "qwen_prompt", None),
+    # ("LlamaForCausalLM", "transformers.models.llama.modeling_llama", "omost_prompt", None),
+    ("T5EncoderModel", "diffsynth.models.flux_text_encoder", "flux_text_encoder_2", "FluxTextEncoder2"),
+    ("CogVideoXTransformer3DModel", "diffsynth.models.cog_dit", "cog_dit", "CogDiT"),
+    ("SiglipModel", "transformers.models.siglip.modeling_siglip", "siglip_vision_model", "SiglipVisionModel"),
+    ("LlamaForCausalLM", "diffsynth.models.hunyuan_video_text_encoder", "hunyuan_video_text_encoder_2", "HunyuanVideoLLMEncoder"),
+    ("Step1Model", "diffsynth.models.stepvideo_text_encoder", "stepvideo_text_encoder_2", "STEP1TextEncoder"),
+]
+patch_model_loader_configs = [
+    # These configs are provided for detecting model type automatically.
+    # The format is (state_dict_keys_hash_with_shape, model_name, model_class, extra_kwargs)
+    ("9a4ab6869ac9b7d6e31f9854e397c867", ["svd_unet"], [SVDUNet], {"add_positional_conv": 128}),
+]
+
+preset_models_on_huggingface = {
+    "HunyuanDiT": [
+        ("Tencent-Hunyuan/HunyuanDiT", "t2i/clip_text_encoder/pytorch_model.bin", "models/HunyuanDiT/t2i/clip_text_encoder"),
+        ("Tencent-Hunyuan/HunyuanDiT", "t2i/mt5/pytorch_model.bin", "models/HunyuanDiT/t2i/mt5"),
+        ("Tencent-Hunyuan/HunyuanDiT", "t2i/model/pytorch_model_ema.pt", "models/HunyuanDiT/t2i/model"),
+        ("Tencent-Hunyuan/HunyuanDiT", "t2i/sdxl-vae-fp16-fix/diffusion_pytorch_model.bin", "models/HunyuanDiT/t2i/sdxl-vae-fp16-fix"),
+    ],
+    "stable-video-diffusion-img2vid-xt": [
+        ("stabilityai/stable-video-diffusion-img2vid-xt", "svd_xt.safetensors", "models/stable_video_diffusion"),
+    ],
+    "ExVideo-SVD-128f-v1": [
+        ("ECNU-CILab/ExVideo-SVD-128f-v1", "model.fp16.safetensors", "models/stable_video_diffusion"),
+    ],
+    # Stable Diffusion
+    "StableDiffusion_v15": [
+        ("benjamin-paine/stable-diffusion-v1-5", "v1-5-pruned-emaonly.safetensors", "models/stable_diffusion"),
+    ],
+    "DreamShaper_8": [
+        ("Yntec/Dreamshaper8", "dreamshaper_8.safetensors", "models/stable_diffusion"),
+    ],
+    # Textual Inversion
+    "TextualInversion_VeryBadImageNegative_v1.3": [
+        ("gemasai/verybadimagenegative_v1.3", "verybadimagenegative_v1.3.pt", "models/textual_inversion"),
+    ],
+    # Stable Diffusion XL
+    "StableDiffusionXL_v1": [
+        ("stabilityai/stable-diffusion-xl-base-1.0", "sd_xl_base_1.0.safetensors", "models/stable_diffusion_xl"),
+    ],
+    "BluePencilXL_v200": [
+        ("frankjoshua/bluePencilXL_v200", "bluePencilXL_v200.safetensors", "models/stable_diffusion_xl"),
+    ],
+    "StableDiffusionXL_Turbo": [
+        ("stabilityai/sdxl-turbo", "sd_xl_turbo_1.0_fp16.safetensors", "models/stable_diffusion_xl_turbo"),
+    ],
+    # Stable Diffusion 3
+    "StableDiffusion3": [
+        ("stabilityai/stable-diffusion-3-medium", "sd3_medium_incl_clips_t5xxlfp16.safetensors", "models/stable_diffusion_3"),
+    ],
+    "StableDiffusion3_without_T5": [
+        ("stabilityai/stable-diffusion-3-medium", "sd3_medium_incl_clips.safetensors", "models/stable_diffusion_3"),
+    ],
+    # ControlNet
+    "ControlNet_v11f1p_sd15_depth": [
+        ("lllyasviel/ControlNet-v1-1", "control_v11f1p_sd15_depth.pth", "models/ControlNet"),
+        ("lllyasviel/Annotators", "dpt_hybrid-midas-501f0c75.pt", "models/Annotators")
+    ],
+    "ControlNet_v11p_sd15_softedge": [
+        ("lllyasviel/ControlNet-v1-1", "control_v11p_sd15_softedge.pth", "models/ControlNet"),
+        ("lllyasviel/Annotators", "ControlNetHED.pth", "models/Annotators")
+    ],
+    "ControlNet_v11f1e_sd15_tile": [
+        ("lllyasviel/ControlNet-v1-1", "control_v11f1e_sd15_tile.pth", "models/ControlNet")
+    ],
+    "ControlNet_v11p_sd15_lineart": [
+        ("lllyasviel/ControlNet-v1-1", "control_v11p_sd15_lineart.pth", "models/ControlNet"),
+        ("lllyasviel/Annotators", "sk_model.pth", "models/Annotators"),
+        ("lllyasviel/Annotators", "sk_model2.pth", "models/Annotators")
+    ],
+    "ControlNet_union_sdxl_promax": [
+        ("xinsir/controlnet-union-sdxl-1.0", "diffusion_pytorch_model_promax.safetensors", "models/ControlNet/controlnet_union"),
+        ("lllyasviel/Annotators", "dpt_hybrid-midas-501f0c75.pt", "models/Annotators")
+    ],
+    # AnimateDiff
+    "AnimateDiff_v2": [
+        ("guoyww/animatediff", "mm_sd_v15_v2.ckpt", "models/AnimateDiff"),
+    ],
+    "AnimateDiff_xl_beta": [
+        ("guoyww/animatediff", "mm_sdxl_v10_beta.ckpt", "models/AnimateDiff"),
+    ],
+
+    # Qwen Prompt
+    "QwenPrompt": [
+        ("Qwen/Qwen2-1.5B-Instruct", "config.json", "models/QwenPrompt/qwen2-1.5b-instruct"),
+        ("Qwen/Qwen2-1.5B-Instruct", "generation_config.json", "models/QwenPrompt/qwen2-1.5b-instruct"),
+        ("Qwen/Qwen2-1.5B-Instruct", "model.safetensors", "models/QwenPrompt/qwen2-1.5b-instruct"),
+        ("Qwen/Qwen2-1.5B-Instruct", "special_tokens_map.json", "models/QwenPrompt/qwen2-1.5b-instruct"),
+        ("Qwen/Qwen2-1.5B-Instruct", "tokenizer.json", "models/QwenPrompt/qwen2-1.5b-instruct"),
+        ("Qwen/Qwen2-1.5B-Instruct", "tokenizer_config.json", "models/QwenPrompt/qwen2-1.5b-instruct"),
+        ("Qwen/Qwen2-1.5B-Instruct", "merges.txt", "models/QwenPrompt/qwen2-1.5b-instruct"),
+        ("Qwen/Qwen2-1.5B-Instruct", "vocab.json", "models/QwenPrompt/qwen2-1.5b-instruct"),
+    ],
+    # Beautiful Prompt
+    "BeautifulPrompt": [
+        ("alibaba-pai/pai-bloom-1b1-text2prompt-sd", "config.json", "models/BeautifulPrompt/pai-bloom-1b1-text2prompt-sd"),
+        ("alibaba-pai/pai-bloom-1b1-text2prompt-sd", "generation_config.json", "models/BeautifulPrompt/pai-bloom-1b1-text2prompt-sd"),
+        ("alibaba-pai/pai-bloom-1b1-text2prompt-sd", "model.safetensors", "models/BeautifulPrompt/pai-bloom-1b1-text2prompt-sd"),
+        ("alibaba-pai/pai-bloom-1b1-text2prompt-sd", "special_tokens_map.json", "models/BeautifulPrompt/pai-bloom-1b1-text2prompt-sd"),
+        ("alibaba-pai/pai-bloom-1b1-text2prompt-sd", "tokenizer.json", "models/BeautifulPrompt/pai-bloom-1b1-text2prompt-sd"),
+        ("alibaba-pai/pai-bloom-1b1-text2prompt-sd", "tokenizer_config.json", "models/BeautifulPrompt/pai-bloom-1b1-text2prompt-sd"),
+    ],
+    # Omost prompt
+    "OmostPrompt":[
+        ("lllyasviel/omost-llama-3-8b-4bits", "model-00001-of-00002.safetensors", "models/OmostPrompt/omost-llama-3-8b-4bits"),
+        ("lllyasviel/omost-llama-3-8b-4bits", "model-00002-of-00002.safetensors", "models/OmostPrompt/omost-llama-3-8b-4bits"),
+        ("lllyasviel/omost-llama-3-8b-4bits", "tokenizer.json", "models/OmostPrompt/omost-llama-3-8b-4bits"),
+        ("lllyasviel/omost-llama-3-8b-4bits", "tokenizer_config.json", "models/OmostPrompt/omost-llama-3-8b-4bits"),  
+        ("lllyasviel/omost-llama-3-8b-4bits", "config.json", "models/OmostPrompt/omost-llama-3-8b-4bits"),
+        ("lllyasviel/omost-llama-3-8b-4bits", "generation_config.json", "models/OmostPrompt/omost-llama-3-8b-4bits"),
+        ("lllyasviel/omost-llama-3-8b-4bits", "model.safetensors.index.json", "models/OmostPrompt/omost-llama-3-8b-4bits"),
+        ("lllyasviel/omost-llama-3-8b-4bits", "special_tokens_map.json", "models/OmostPrompt/omost-llama-3-8b-4bits"),
+    ],
+    # Translator
+    "opus-mt-zh-en": [
+        ("Helsinki-NLP/opus-mt-zh-en", "config.json", "models/translator/opus-mt-zh-en"),
+        ("Helsinki-NLP/opus-mt-zh-en", "generation_config.json", "models/translator/opus-mt-zh-en"),
+        ("Helsinki-NLP/opus-mt-zh-en", "metadata.json", "models/translator/opus-mt-zh-en"),
+        ("Helsinki-NLP/opus-mt-zh-en", "pytorch_model.bin", "models/translator/opus-mt-zh-en"),
+        ("Helsinki-NLP/opus-mt-zh-en", "source.spm", "models/translator/opus-mt-zh-en"),
+        ("Helsinki-NLP/opus-mt-zh-en", "target.spm", "models/translator/opus-mt-zh-en"),
+        ("Helsinki-NLP/opus-mt-zh-en", "tokenizer_config.json", "models/translator/opus-mt-zh-en"),
+        ("Helsinki-NLP/opus-mt-zh-en", "vocab.json", "models/translator/opus-mt-zh-en"),
+    ],
+    # IP-Adapter
+    "IP-Adapter-SD": [
+        ("h94/IP-Adapter", "models/image_encoder/model.safetensors", "models/IpAdapter/stable_diffusion/image_encoder"),
+        ("h94/IP-Adapter", "models/ip-adapter_sd15.bin", "models/IpAdapter/stable_diffusion"),
+    ],
+    "IP-Adapter-SDXL": [
+        ("h94/IP-Adapter", "sdxl_models/image_encoder/model.safetensors", "models/IpAdapter/stable_diffusion_xl/image_encoder"),
+        ("h94/IP-Adapter", "sdxl_models/ip-adapter_sdxl.bin", "models/IpAdapter/stable_diffusion_xl"),
+    ],
+    "SDXL-vae-fp16-fix": [
+        ("madebyollin/sdxl-vae-fp16-fix", "diffusion_pytorch_model.safetensors", "models/sdxl-vae-fp16-fix")
+    ],
+    # Kolors
+    "Kolors": [
+        ("Kwai-Kolors/Kolors", "text_encoder/config.json", "models/kolors/Kolors/text_encoder"),
+        ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model.bin.index.json", "models/kolors/Kolors/text_encoder"),
+        ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model-00001-of-00007.bin", "models/kolors/Kolors/text_encoder"),
+        ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model-00002-of-00007.bin", "models/kolors/Kolors/text_encoder"),
+        ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model-00003-of-00007.bin", "models/kolors/Kolors/text_encoder"),
+        ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model-00004-of-00007.bin", "models/kolors/Kolors/text_encoder"),
+        ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model-00005-of-00007.bin", "models/kolors/Kolors/text_encoder"),
+        ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model-00006-of-00007.bin", "models/kolors/Kolors/text_encoder"),
+        ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model-00007-of-00007.bin", "models/kolors/Kolors/text_encoder"),
+        ("Kwai-Kolors/Kolors", "unet/diffusion_pytorch_model.safetensors", "models/kolors/Kolors/unet"),
+        ("Kwai-Kolors/Kolors", "vae/diffusion_pytorch_model.safetensors", "models/kolors/Kolors/vae"),
+    ],
+    # FLUX
+    "FLUX.1-dev": [
+        ("black-forest-labs/FLUX.1-dev", "text_encoder/model.safetensors", "models/FLUX/FLUX.1-dev/text_encoder"),
+        ("black-forest-labs/FLUX.1-dev", "text_encoder_2/config.json", "models/FLUX/FLUX.1-dev/text_encoder_2"),
+        ("black-forest-labs/FLUX.1-dev", "text_encoder_2/model-00001-of-00002.safetensors", "models/FLUX/FLUX.1-dev/text_encoder_2"),
+        ("black-forest-labs/FLUX.1-dev", "text_encoder_2/model-00002-of-00002.safetensors", "models/FLUX/FLUX.1-dev/text_encoder_2"),
+        ("black-forest-labs/FLUX.1-dev", "text_encoder_2/model.safetensors.index.json", "models/FLUX/FLUX.1-dev/text_encoder_2"),
+        ("black-forest-labs/FLUX.1-dev", "ae.safetensors", "models/FLUX/FLUX.1-dev"),
+        ("black-forest-labs/FLUX.1-dev", "flux1-dev.safetensors", "models/FLUX/FLUX.1-dev"),
+    ],
+    "InstantX/FLUX.1-dev-IP-Adapter": {
+        "file_list": [
+            ("InstantX/FLUX.1-dev-IP-Adapter", "ip-adapter.bin", "models/IpAdapter/InstantX/FLUX.1-dev-IP-Adapter"),
+            ("google/siglip-so400m-patch14-384", "model.safetensors", "models/IpAdapter/InstantX/FLUX.1-dev-IP-Adapter/image_encoder"),
+            ("google/siglip-so400m-patch14-384", "config.json", "models/IpAdapter/InstantX/FLUX.1-dev-IP-Adapter/image_encoder"),
+        ],
+        "load_path": [
+            "models/IpAdapter/InstantX/FLUX.1-dev-IP-Adapter/ip-adapter.bin",
+            "models/IpAdapter/InstantX/FLUX.1-dev-IP-Adapter/image_encoder",
+        ],
+    },
+    # RIFE
+    "RIFE": [
+        ("AlexWortega/RIFE", "flownet.pkl", "models/RIFE"),
+    ],
+    # CogVideo
+    "CogVideoX-5B": [
+        ("THUDM/CogVideoX-5b", "text_encoder/config.json", "models/CogVideo/CogVideoX-5b/text_encoder"),
+        ("THUDM/CogVideoX-5b", "text_encoder/model.safetensors.index.json", "models/CogVideo/CogVideoX-5b/text_encoder"),
+        ("THUDM/CogVideoX-5b", "text_encoder/model-00001-of-00002.safetensors", "models/CogVideo/CogVideoX-5b/text_encoder"),
+        ("THUDM/CogVideoX-5b", "text_encoder/model-00002-of-00002.safetensors", "models/CogVideo/CogVideoX-5b/text_encoder"),
+        ("THUDM/CogVideoX-5b", "transformer/config.json", "models/CogVideo/CogVideoX-5b/transformer"),
+        ("THUDM/CogVideoX-5b", "transformer/diffusion_pytorch_model.safetensors.index.json", "models/CogVideo/CogVideoX-5b/transformer"),
+        ("THUDM/CogVideoX-5b", "transformer/diffusion_pytorch_model-00001-of-00002.safetensors", "models/CogVideo/CogVideoX-5b/transformer"),
+        ("THUDM/CogVideoX-5b", "transformer/diffusion_pytorch_model-00002-of-00002.safetensors", "models/CogVideo/CogVideoX-5b/transformer"),
+        ("THUDM/CogVideoX-5b", "vae/diffusion_pytorch_model.safetensors", "models/CogVideo/CogVideoX-5b/vae"),
+    ],
+    # Stable Diffusion 3.5
+    "StableDiffusion3.5-large": [
+        ("stabilityai/stable-diffusion-3.5-large", "sd3.5_large.safetensors", "models/stable_diffusion_3"),
+        ("stabilityai/stable-diffusion-3.5-large", "text_encoders/clip_l.safetensors", "models/stable_diffusion_3/text_encoders"),
+        ("stabilityai/stable-diffusion-3.5-large", "text_encoders/clip_g.safetensors", "models/stable_diffusion_3/text_encoders"),
+        ("stabilityai/stable-diffusion-3.5-large", "text_encoders/t5xxl_fp16.safetensors", "models/stable_diffusion_3/text_encoders"),
+    ],
+}
+preset_models_on_modelscope = {
+    # Hunyuan DiT
+    "HunyuanDiT": [
+        ("modelscope/HunyuanDiT", "t2i/clip_text_encoder/pytorch_model.bin", "models/HunyuanDiT/t2i/clip_text_encoder"),
+        ("modelscope/HunyuanDiT", "t2i/mt5/pytorch_model.bin", "models/HunyuanDiT/t2i/mt5"),
+        ("modelscope/HunyuanDiT", "t2i/model/pytorch_model_ema.pt", "models/HunyuanDiT/t2i/model"),
+        ("modelscope/HunyuanDiT", "t2i/sdxl-vae-fp16-fix/diffusion_pytorch_model.bin", "models/HunyuanDiT/t2i/sdxl-vae-fp16-fix"),
+    ],
+    # Stable Video Diffusion
+    "stable-video-diffusion-img2vid-xt": [
+        ("AI-ModelScope/stable-video-diffusion-img2vid-xt", "svd_xt.safetensors", "models/stable_video_diffusion"),
+    ],
+    # ExVideo
+    "ExVideo-SVD-128f-v1": [
+        ("ECNU-CILab/ExVideo-SVD-128f-v1", "model.fp16.safetensors", "models/stable_video_diffusion"),
+    ],
+    "ExVideo-CogVideoX-LoRA-129f-v1": [
+        ("ECNU-CILab/ExVideo-CogVideoX-LoRA-129f-v1", "ExVideo-CogVideoX-LoRA-129f-v1.safetensors", "models/lora"),
+    ],
+    # Stable Diffusion
+    "StableDiffusion_v15": [
+        ("AI-ModelScope/stable-diffusion-v1-5", "v1-5-pruned-emaonly.safetensors", "models/stable_diffusion"),
+    ],
+    "DreamShaper_8": [
+        ("sd_lora/dreamshaper_8", "dreamshaper_8.safetensors", "models/stable_diffusion"),
+    ],
+    "AingDiffusion_v12": [
+        ("sd_lora/aingdiffusion_v12", "aingdiffusion_v12.safetensors", "models/stable_diffusion"),
+    ],
+    "Flat2DAnimerge_v45Sharp": [
+        ("sd_lora/Flat-2D-Animerge", "flat2DAnimerge_v45Sharp.safetensors", "models/stable_diffusion"),
+    ],
+    # Textual Inversion
+    "TextualInversion_VeryBadImageNegative_v1.3": [
+        ("sd_lora/verybadimagenegative_v1.3", "verybadimagenegative_v1.3.pt", "models/textual_inversion"),
+    ],
+    # Stable Diffusion XL
+    "StableDiffusionXL_v1": [
+        ("AI-ModelScope/stable-diffusion-xl-base-1.0", "sd_xl_base_1.0.safetensors", "models/stable_diffusion_xl"),
+    ],
+    "BluePencilXL_v200": [
+        ("sd_lora/bluePencilXL_v200", "bluePencilXL_v200.safetensors", "models/stable_diffusion_xl"),
+    ],
+    "StableDiffusionXL_Turbo": [
+        ("AI-ModelScope/sdxl-turbo", "sd_xl_turbo_1.0_fp16.safetensors", "models/stable_diffusion_xl_turbo"),
+    ],
+    "SDXL_lora_zyd232_ChineseInkStyle_SDXL_v1_0": [
+        ("sd_lora/zyd232_ChineseInkStyle_SDXL_v1_0", "zyd232_ChineseInkStyle_SDXL_v1_0.safetensors", "models/lora"),
+    ],
+    # Stable Diffusion 3
+    "StableDiffusion3": [
+        ("AI-ModelScope/stable-diffusion-3-medium", "sd3_medium_incl_clips_t5xxlfp16.safetensors", "models/stable_diffusion_3"),
+    ],
+    "StableDiffusion3_without_T5": [
+        ("AI-ModelScope/stable-diffusion-3-medium", "sd3_medium_incl_clips.safetensors", "models/stable_diffusion_3"),
+    ],
+    # ControlNet
+    "ControlNet_v11f1p_sd15_depth": [
+        ("AI-ModelScope/ControlNet-v1-1", "control_v11f1p_sd15_depth.pth", "models/ControlNet"),
+        ("sd_lora/Annotators", "dpt_hybrid-midas-501f0c75.pt", "models/Annotators")
+    ],
+    "ControlNet_v11p_sd15_softedge": [
+        ("AI-ModelScope/ControlNet-v1-1", "control_v11p_sd15_softedge.pth", "models/ControlNet"),
+        ("sd_lora/Annotators", "ControlNetHED.pth", "models/Annotators")
+    ],
+    "ControlNet_v11f1e_sd15_tile": [
+        ("AI-ModelScope/ControlNet-v1-1", "control_v11f1e_sd15_tile.pth", "models/ControlNet")
+    ],
+    "ControlNet_v11p_sd15_lineart": [
+        ("AI-ModelScope/ControlNet-v1-1", "control_v11p_sd15_lineart.pth", "models/ControlNet"),
+        ("sd_lora/Annotators", "sk_model.pth", "models/Annotators"),
+        ("sd_lora/Annotators", "sk_model2.pth", "models/Annotators")
+    ],
+    "ControlNet_union_sdxl_promax": [
+        ("AI-ModelScope/controlnet-union-sdxl-1.0", "diffusion_pytorch_model_promax.safetensors", "models/ControlNet/controlnet_union"),
+        ("sd_lora/Annotators", "dpt_hybrid-midas-501f0c75.pt", "models/Annotators")
+    ],
+    "Annotators:Depth": [
+        ("sd_lora/Annotators", "dpt_hybrid-midas-501f0c75.pt", "models/Annotators"),
+    ],
+    "Annotators:Softedge": [
+        ("sd_lora/Annotators", "ControlNetHED.pth", "models/Annotators"),
+    ],
+    "Annotators:Lineart": [
+        ("sd_lora/Annotators", "sk_model.pth", "models/Annotators"),
+        ("sd_lora/Annotators", "sk_model2.pth", "models/Annotators"),
+    ],
+    "Annotators:Normal": [
+        ("sd_lora/Annotators", "scannet.pt", "models/Annotators"),
+    ],
+    "Annotators:Openpose": [
+        ("sd_lora/Annotators", "body_pose_model.pth", "models/Annotators"),
+        ("sd_lora/Annotators", "facenet.pth", "models/Annotators"),
+        ("sd_lora/Annotators", "hand_pose_model.pth", "models/Annotators"),
+    ],
+    # AnimateDiff
+    "AnimateDiff_v2": [
+        ("Shanghai_AI_Laboratory/animatediff", "mm_sd_v15_v2.ckpt", "models/AnimateDiff"),
+    ],
+    "AnimateDiff_xl_beta": [
+        ("Shanghai_AI_Laboratory/animatediff", "mm_sdxl_v10_beta.ckpt", "models/AnimateDiff"),
+    ],
+    # RIFE
+    "RIFE": [
+        ("Damo_XR_Lab/cv_rife_video-frame-interpolation", "flownet.pkl", "models/RIFE"),
+    ],
+    # Qwen Prompt
+    "QwenPrompt": {
+        "file_list": [
+            ("qwen/Qwen2-1.5B-Instruct", "config.json", "models/QwenPrompt/qwen2-1.5b-instruct"),
+            ("qwen/Qwen2-1.5B-Instruct", "generation_config.json", "models/QwenPrompt/qwen2-1.5b-instruct"),
+            ("qwen/Qwen2-1.5B-Instruct", "model.safetensors", "models/QwenPrompt/qwen2-1.5b-instruct"),
+            ("qwen/Qwen2-1.5B-Instruct", "special_tokens_map.json", "models/QwenPrompt/qwen2-1.5b-instruct"),
+            ("qwen/Qwen2-1.5B-Instruct", "tokenizer.json", "models/QwenPrompt/qwen2-1.5b-instruct"),
+            ("qwen/Qwen2-1.5B-Instruct", "tokenizer_config.json", "models/QwenPrompt/qwen2-1.5b-instruct"),
+            ("qwen/Qwen2-1.5B-Instruct", "merges.txt", "models/QwenPrompt/qwen2-1.5b-instruct"),
+            ("qwen/Qwen2-1.5B-Instruct", "vocab.json", "models/QwenPrompt/qwen2-1.5b-instruct"),
+        ],
+        "load_path": [
+            "models/QwenPrompt/qwen2-1.5b-instruct",
+        ],
+    },
+    # Beautiful Prompt
+    "BeautifulPrompt": {
+        "file_list": [
+            ("AI-ModelScope/pai-bloom-1b1-text2prompt-sd", "config.json", "models/BeautifulPrompt/pai-bloom-1b1-text2prompt-sd"),
+            ("AI-ModelScope/pai-bloom-1b1-text2prompt-sd", "generation_config.json", "models/BeautifulPrompt/pai-bloom-1b1-text2prompt-sd"),
+            ("AI-ModelScope/pai-bloom-1b1-text2prompt-sd", "model.safetensors", "models/BeautifulPrompt/pai-bloom-1b1-text2prompt-sd"),
+            ("AI-ModelScope/pai-bloom-1b1-text2prompt-sd", "special_tokens_map.json", "models/BeautifulPrompt/pai-bloom-1b1-text2prompt-sd"),
+            ("AI-ModelScope/pai-bloom-1b1-text2prompt-sd", "tokenizer.json", "models/BeautifulPrompt/pai-bloom-1b1-text2prompt-sd"),
+            ("AI-ModelScope/pai-bloom-1b1-text2prompt-sd", "tokenizer_config.json", "models/BeautifulPrompt/pai-bloom-1b1-text2prompt-sd"),
+        ],
+        "load_path": [
+            "models/BeautifulPrompt/pai-bloom-1b1-text2prompt-sd",
+        ],
+    },
+    # Omost prompt
+    "OmostPrompt": {
+        "file_list": [
+            ("Omost/omost-llama-3-8b-4bits", "model-00001-of-00002.safetensors", "models/OmostPrompt/omost-llama-3-8b-4bits"),
+            ("Omost/omost-llama-3-8b-4bits", "model-00002-of-00002.safetensors", "models/OmostPrompt/omost-llama-3-8b-4bits"),
+            ("Omost/omost-llama-3-8b-4bits", "tokenizer.json", "models/OmostPrompt/omost-llama-3-8b-4bits"),
+            ("Omost/omost-llama-3-8b-4bits", "tokenizer_config.json", "models/OmostPrompt/omost-llama-3-8b-4bits"),  
+            ("Omost/omost-llama-3-8b-4bits", "config.json", "models/OmostPrompt/omost-llama-3-8b-4bits"),
+            ("Omost/omost-llama-3-8b-4bits", "generation_config.json", "models/OmostPrompt/omost-llama-3-8b-4bits"),
+            ("Omost/omost-llama-3-8b-4bits", "model.safetensors.index.json", "models/OmostPrompt/omost-llama-3-8b-4bits"),
+            ("Omost/omost-llama-3-8b-4bits", "special_tokens_map.json", "models/OmostPrompt/omost-llama-3-8b-4bits"),
+        ],
+        "load_path": [
+            "models/OmostPrompt/omost-llama-3-8b-4bits",
+        ],
+    },
+    # Translator
+    "opus-mt-zh-en": {
+        "file_list": [
+            ("moxying/opus-mt-zh-en", "config.json", "models/translator/opus-mt-zh-en"),
+            ("moxying/opus-mt-zh-en", "generation_config.json", "models/translator/opus-mt-zh-en"),
+            ("moxying/opus-mt-zh-en", "metadata.json", "models/translator/opus-mt-zh-en"),
+            ("moxying/opus-mt-zh-en", "pytorch_model.bin", "models/translator/opus-mt-zh-en"),
+            ("moxying/opus-mt-zh-en", "source.spm", "models/translator/opus-mt-zh-en"),
+            ("moxying/opus-mt-zh-en", "target.spm", "models/translator/opus-mt-zh-en"),
+            ("moxying/opus-mt-zh-en", "tokenizer_config.json", "models/translator/opus-mt-zh-en"),
+            ("moxying/opus-mt-zh-en", "vocab.json", "models/translator/opus-mt-zh-en"),
+        ],
+        "load_path": [
+            "models/translator/opus-mt-zh-en",
+        ],
+    },
+    # IP-Adapter
+    "IP-Adapter-SD": [
+        ("AI-ModelScope/IP-Adapter", "models/image_encoder/model.safetensors", "models/IpAdapter/stable_diffusion/image_encoder"),
+        ("AI-ModelScope/IP-Adapter", "models/ip-adapter_sd15.bin", "models/IpAdapter/stable_diffusion"),
+    ],
+    "IP-Adapter-SDXL": [
+        ("AI-ModelScope/IP-Adapter", "sdxl_models/image_encoder/model.safetensors", "models/IpAdapter/stable_diffusion_xl/image_encoder"),
+        ("AI-ModelScope/IP-Adapter", "sdxl_models/ip-adapter_sdxl.bin", "models/IpAdapter/stable_diffusion_xl"),
+    ],
+    # Kolors
+    "Kolors": {
+        "file_list": [
+            ("Kwai-Kolors/Kolors", "text_encoder/config.json", "models/kolors/Kolors/text_encoder"),
+            ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model.bin.index.json", "models/kolors/Kolors/text_encoder"),
+            ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model-00001-of-00007.bin", "models/kolors/Kolors/text_encoder"),
+            ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model-00002-of-00007.bin", "models/kolors/Kolors/text_encoder"),
+            ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model-00003-of-00007.bin", "models/kolors/Kolors/text_encoder"),
+            ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model-00004-of-00007.bin", "models/kolors/Kolors/text_encoder"),
+            ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model-00005-of-00007.bin", "models/kolors/Kolors/text_encoder"),
+            ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model-00006-of-00007.bin", "models/kolors/Kolors/text_encoder"),
+            ("Kwai-Kolors/Kolors", "text_encoder/pytorch_model-00007-of-00007.bin", "models/kolors/Kolors/text_encoder"),
+            ("Kwai-Kolors/Kolors", "unet/diffusion_pytorch_model.safetensors", "models/kolors/Kolors/unet"),
+            ("Kwai-Kolors/Kolors", "vae/diffusion_pytorch_model.safetensors", "models/kolors/Kolors/vae"),
+        ],
+        "load_path": [
+            "models/kolors/Kolors/text_encoder",
+            "models/kolors/Kolors/unet/diffusion_pytorch_model.safetensors",
+            "models/kolors/Kolors/vae/diffusion_pytorch_model.safetensors",
+        ],
+    },
+    "SDXL-vae-fp16-fix": [
+        ("AI-ModelScope/sdxl-vae-fp16-fix", "diffusion_pytorch_model.safetensors", "models/sdxl-vae-fp16-fix")
+    ],
+    # FLUX
+    "FLUX.1-dev": {
+        "file_list": [
+            ("AI-ModelScope/FLUX.1-dev", "text_encoder/model.safetensors", "models/FLUX/FLUX.1-dev/text_encoder"),
+            ("AI-ModelScope/FLUX.1-dev", "text_encoder_2/config.json", "models/FLUX/FLUX.1-dev/text_encoder_2"),
+            ("AI-ModelScope/FLUX.1-dev", "text_encoder_2/model-00001-of-00002.safetensors", "models/FLUX/FLUX.1-dev/text_encoder_2"),
+            ("AI-ModelScope/FLUX.1-dev", "text_encoder_2/model-00002-of-00002.safetensors", "models/FLUX/FLUX.1-dev/text_encoder_2"),
+            ("AI-ModelScope/FLUX.1-dev", "text_encoder_2/model.safetensors.index.json", "models/FLUX/FLUX.1-dev/text_encoder_2"),
+            ("AI-ModelScope/FLUX.1-dev", "ae.safetensors", "models/FLUX/FLUX.1-dev"),
+            ("AI-ModelScope/FLUX.1-dev", "flux1-dev.safetensors", "models/FLUX/FLUX.1-dev"),
+        ],
+        "load_path": [
+            "models/FLUX/FLUX.1-dev/text_encoder/model.safetensors",
+            "models/FLUX/FLUX.1-dev/text_encoder_2",
+            "models/FLUX/FLUX.1-dev/ae.safetensors",
+            "models/FLUX/FLUX.1-dev/flux1-dev.safetensors"
+        ],
+    },
+    "FLUX.1-schnell": {
+        "file_list": [
+            ("AI-ModelScope/FLUX.1-dev", "text_encoder/model.safetensors", "models/FLUX/FLUX.1-dev/text_encoder"),
+            ("AI-ModelScope/FLUX.1-dev", "text_encoder_2/config.json", "models/FLUX/FLUX.1-dev/text_encoder_2"),
+            ("AI-ModelScope/FLUX.1-dev", "text_encoder_2/model-00001-of-00002.safetensors", "models/FLUX/FLUX.1-dev/text_encoder_2"),
+            ("AI-ModelScope/FLUX.1-dev", "text_encoder_2/model-00002-of-00002.safetensors", "models/FLUX/FLUX.1-dev/text_encoder_2"),
+            ("AI-ModelScope/FLUX.1-dev", "text_encoder_2/model.safetensors.index.json", "models/FLUX/FLUX.1-dev/text_encoder_2"),
+            ("AI-ModelScope/FLUX.1-dev", "ae.safetensors", "models/FLUX/FLUX.1-dev"),
+            ("AI-ModelScope/FLUX.1-schnell", "flux1-schnell.safetensors", "models/FLUX/FLUX.1-schnell"),
+        ],
+        "load_path": [
+            "models/FLUX/FLUX.1-dev/text_encoder/model.safetensors",
+            "models/FLUX/FLUX.1-dev/text_encoder_2",
+            "models/FLUX/FLUX.1-dev/ae.safetensors",
+            "models/FLUX/FLUX.1-schnell/flux1-schnell.safetensors"
+        ],
+    },
+    "InstantX/FLUX.1-dev-Controlnet-Union-alpha": [
+        ("InstantX/FLUX.1-dev-Controlnet-Union-alpha", "diffusion_pytorch_model.safetensors", "models/ControlNet/InstantX/FLUX.1-dev-Controlnet-Union-alpha"),
+    ],
+    "jasperai/Flux.1-dev-Controlnet-Depth": [
+        ("jasperai/Flux.1-dev-Controlnet-Depth", "diffusion_pytorch_model.safetensors", "models/ControlNet/jasperai/Flux.1-dev-Controlnet-Depth"),
+    ],
+    "jasperai/Flux.1-dev-Controlnet-Surface-Normals": [
+        ("jasperai/Flux.1-dev-Controlnet-Surface-Normals", "diffusion_pytorch_model.safetensors", "models/ControlNet/jasperai/Flux.1-dev-Controlnet-Surface-Normals"),
+    ],
+    "jasperai/Flux.1-dev-Controlnet-Upscaler": [
+        ("jasperai/Flux.1-dev-Controlnet-Upscaler", "diffusion_pytorch_model.safetensors", "models/ControlNet/jasperai/Flux.1-dev-Controlnet-Upscaler"),
+    ],
+    "alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Alpha": [
+        ("alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Alpha", "diffusion_pytorch_model.safetensors", "models/ControlNet/alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Alpha"),
+    ],
+    "alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Beta": [
+        ("alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Beta", "diffusion_pytorch_model.safetensors", "models/ControlNet/alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Beta"),
+    ],
+    "Shakker-Labs/FLUX.1-dev-ControlNet-Depth": [
+        ("Shakker-Labs/FLUX.1-dev-ControlNet-Depth", "diffusion_pytorch_model.safetensors", "models/ControlNet/Shakker-Labs/FLUX.1-dev-ControlNet-Depth"),
+    ],
+    "Shakker-Labs/FLUX.1-dev-ControlNet-Union-Pro": [
+        ("Shakker-Labs/FLUX.1-dev-ControlNet-Union-Pro", "diffusion_pytorch_model.safetensors", "models/ControlNet/Shakker-Labs/FLUX.1-dev-ControlNet-Union-Pro"),
+    ],
+    "InstantX/FLUX.1-dev-IP-Adapter": {
+        "file_list": [
+            ("InstantX/FLUX.1-dev-IP-Adapter", "ip-adapter.bin", "models/IpAdapter/InstantX/FLUX.1-dev-IP-Adapter"),
+            ("AI-ModelScope/siglip-so400m-patch14-384", "model.safetensors", "models/IpAdapter/InstantX/FLUX.1-dev-IP-Adapter/image_encoder"),
+            ("AI-ModelScope/siglip-so400m-patch14-384", "config.json", "models/IpAdapter/InstantX/FLUX.1-dev-IP-Adapter/image_encoder"),
+        ],
+        "load_path": [
+            "models/IpAdapter/InstantX/FLUX.1-dev-IP-Adapter/ip-adapter.bin",
+            "models/IpAdapter/InstantX/FLUX.1-dev-IP-Adapter/image_encoder",
+        ],
+    },
+    # ESRGAN
+    "ESRGAN_x4": [
+        ("AI-ModelScope/Real-ESRGAN", "RealESRGAN_x4.pth", "models/ESRGAN"),
+    ],
+    # RIFE
+    "RIFE": [
+        ("AI-ModelScope/RIFE", "flownet.pkl", "models/RIFE"),
+    ],
+    # Omnigen
+    "OmniGen-v1": {
+        "file_list": [
+            ("BAAI/OmniGen-v1", "vae/diffusion_pytorch_model.safetensors", "models/OmniGen/OmniGen-v1/vae"),
+            ("BAAI/OmniGen-v1", "model.safetensors", "models/OmniGen/OmniGen-v1"),
+            ("BAAI/OmniGen-v1", "config.json", "models/OmniGen/OmniGen-v1"),
+            ("BAAI/OmniGen-v1", "special_tokens_map.json", "models/OmniGen/OmniGen-v1"),
+            ("BAAI/OmniGen-v1", "tokenizer_config.json", "models/OmniGen/OmniGen-v1"),
+            ("BAAI/OmniGen-v1", "tokenizer.json", "models/OmniGen/OmniGen-v1"),
+        ],
+        "load_path": [
+            "models/OmniGen/OmniGen-v1/vae/diffusion_pytorch_model.safetensors",
+            "models/OmniGen/OmniGen-v1/model.safetensors",
+        ]
+    },
+    # CogVideo
+    "CogVideoX-5B": {
+        "file_list": [
+            ("ZhipuAI/CogVideoX-5b", "text_encoder/config.json", "models/CogVideo/CogVideoX-5b/text_encoder"),
+            ("ZhipuAI/CogVideoX-5b", "text_encoder/model.safetensors.index.json", "models/CogVideo/CogVideoX-5b/text_encoder"),
+            ("ZhipuAI/CogVideoX-5b", "text_encoder/model-00001-of-00002.safetensors", "models/CogVideo/CogVideoX-5b/text_encoder"),
+            ("ZhipuAI/CogVideoX-5b", "text_encoder/model-00002-of-00002.safetensors", "models/CogVideo/CogVideoX-5b/text_encoder"),
+            ("ZhipuAI/CogVideoX-5b", "transformer/config.json", "models/CogVideo/CogVideoX-5b/transformer"),
+            ("ZhipuAI/CogVideoX-5b", "transformer/diffusion_pytorch_model.safetensors.index.json", "models/CogVideo/CogVideoX-5b/transformer"),
+            ("ZhipuAI/CogVideoX-5b", "transformer/diffusion_pytorch_model-00001-of-00002.safetensors", "models/CogVideo/CogVideoX-5b/transformer"),
+            ("ZhipuAI/CogVideoX-5b", "transformer/diffusion_pytorch_model-00002-of-00002.safetensors", "models/CogVideo/CogVideoX-5b/transformer"),
+            ("ZhipuAI/CogVideoX-5b", "vae/diffusion_pytorch_model.safetensors", "models/CogVideo/CogVideoX-5b/vae"),
+        ],
+        "load_path": [
+            "models/CogVideo/CogVideoX-5b/text_encoder",
+            "models/CogVideo/CogVideoX-5b/transformer",
+            "models/CogVideo/CogVideoX-5b/vae/diffusion_pytorch_model.safetensors",
+        ],
+    },
+    # Stable Diffusion 3.5
+    "StableDiffusion3.5-large": [
+        ("AI-ModelScope/stable-diffusion-3.5-large", "sd3.5_large.safetensors", "models/stable_diffusion_3"),
+        ("AI-ModelScope/stable-diffusion-3.5-large", "text_encoders/clip_l.safetensors", "models/stable_diffusion_3/text_encoders"),
+        ("AI-ModelScope/stable-diffusion-3.5-large", "text_encoders/clip_g.safetensors", "models/stable_diffusion_3/text_encoders"),
+        ("AI-ModelScope/stable-diffusion-3.5-large", "text_encoders/t5xxl_fp16.safetensors", "models/stable_diffusion_3/text_encoders"),
+    ],
+    "StableDiffusion3.5-medium": [
+        ("AI-ModelScope/stable-diffusion-3.5-medium", "sd3.5_medium.safetensors", "models/stable_diffusion_3"),
+        ("AI-ModelScope/stable-diffusion-3.5-large", "text_encoders/clip_l.safetensors", "models/stable_diffusion_3/text_encoders"),
+        ("AI-ModelScope/stable-diffusion-3.5-large", "text_encoders/clip_g.safetensors", "models/stable_diffusion_3/text_encoders"),
+        ("AI-ModelScope/stable-diffusion-3.5-large", "text_encoders/t5xxl_fp16.safetensors", "models/stable_diffusion_3/text_encoders"),
+    ],
+    "StableDiffusion3.5-large-turbo": [
+        ("AI-ModelScope/stable-diffusion-3.5-large-turbo", "sd3.5_large_turbo.safetensors", "models/stable_diffusion_3"),
+        ("AI-ModelScope/stable-diffusion-3.5-large", "text_encoders/clip_l.safetensors", "models/stable_diffusion_3/text_encoders"),
+        ("AI-ModelScope/stable-diffusion-3.5-large", "text_encoders/clip_g.safetensors", "models/stable_diffusion_3/text_encoders"),
+        ("AI-ModelScope/stable-diffusion-3.5-large", "text_encoders/t5xxl_fp16.safetensors", "models/stable_diffusion_3/text_encoders"),
+    ],
+    "HunyuanVideo":{
+        "file_list": [
+            ("AI-ModelScope/clip-vit-large-patch14", "model.safetensors", "models/HunyuanVideo/text_encoder"),
+            ("DiffSynth-Studio/HunyuanVideo_MLLM_text_encoder", "model-00001-of-00004.safetensors", "models/HunyuanVideo/text_encoder_2"),
+            ("DiffSynth-Studio/HunyuanVideo_MLLM_text_encoder", "model-00002-of-00004.safetensors", "models/HunyuanVideo/text_encoder_2"),
+            ("DiffSynth-Studio/HunyuanVideo_MLLM_text_encoder", "model-00003-of-00004.safetensors", "models/HunyuanVideo/text_encoder_2"),
+            ("DiffSynth-Studio/HunyuanVideo_MLLM_text_encoder", "model-00004-of-00004.safetensors", "models/HunyuanVideo/text_encoder_2"),
+            ("DiffSynth-Studio/HunyuanVideo_MLLM_text_encoder", "config.json", "models/HunyuanVideo/text_encoder_2"),
+            ("DiffSynth-Studio/HunyuanVideo_MLLM_text_encoder", "model.safetensors.index.json", "models/HunyuanVideo/text_encoder_2"),
+            ("AI-ModelScope/HunyuanVideo", "hunyuan-video-t2v-720p/vae/pytorch_model.pt", "models/HunyuanVideo/vae"),
+            ("AI-ModelScope/HunyuanVideo", "hunyuan-video-t2v-720p/transformers/mp_rank_00_model_states.pt", "models/HunyuanVideo/transformers")
+        ],
+        "load_path": [
+            "models/HunyuanVideo/text_encoder/model.safetensors",
+            "models/HunyuanVideo/text_encoder_2",
+            "models/HunyuanVideo/vae/pytorch_model.pt",
+            "models/HunyuanVideo/transformers/mp_rank_00_model_states.pt"
+        ],
+    },
+    "HunyuanVideo-fp8":{
+        "file_list": [
+            ("AI-ModelScope/clip-vit-large-patch14", "model.safetensors", "models/HunyuanVideo/text_encoder"),
+            ("DiffSynth-Studio/HunyuanVideo_MLLM_text_encoder", "model-00001-of-00004.safetensors", "models/HunyuanVideo/text_encoder_2"),
+            ("DiffSynth-Studio/HunyuanVideo_MLLM_text_encoder", "model-00002-of-00004.safetensors", "models/HunyuanVideo/text_encoder_2"),
+            ("DiffSynth-Studio/HunyuanVideo_MLLM_text_encoder", "model-00003-of-00004.safetensors", "models/HunyuanVideo/text_encoder_2"),
+            ("DiffSynth-Studio/HunyuanVideo_MLLM_text_encoder", "model-00004-of-00004.safetensors", "models/HunyuanVideo/text_encoder_2"),
+            ("DiffSynth-Studio/HunyuanVideo_MLLM_text_encoder", "config.json", "models/HunyuanVideo/text_encoder_2"),
+            ("DiffSynth-Studio/HunyuanVideo_MLLM_text_encoder", "model.safetensors.index.json", "models/HunyuanVideo/text_encoder_2"),
+            ("AI-ModelScope/HunyuanVideo", "hunyuan-video-t2v-720p/vae/pytorch_model.pt", "models/HunyuanVideo/vae"),
+            ("DiffSynth-Studio/HunyuanVideo-safetensors", "model.fp8.safetensors", "models/HunyuanVideo/transformers")
+        ],
+        "load_path": [
+            "models/HunyuanVideo/text_encoder/model.safetensors",
+            "models/HunyuanVideo/text_encoder_2",
+            "models/HunyuanVideo/vae/pytorch_model.pt",
+            "models/HunyuanVideo/transformers/model.fp8.safetensors"
+        ],
+    },
+}
+Preset_model_id: TypeAlias = Literal[
+    "HunyuanDiT",
+    "stable-video-diffusion-img2vid-xt",
+    "ExVideo-SVD-128f-v1",
+    "ExVideo-CogVideoX-LoRA-129f-v1",
+    "StableDiffusion_v15",
+    "DreamShaper_8",
+    "AingDiffusion_v12",
+    "Flat2DAnimerge_v45Sharp",
+    "TextualInversion_VeryBadImageNegative_v1.3",
+    "StableDiffusionXL_v1",
+    "BluePencilXL_v200",
+    "StableDiffusionXL_Turbo",
+    "ControlNet_v11f1p_sd15_depth",
+    "ControlNet_v11p_sd15_softedge",
+    "ControlNet_v11f1e_sd15_tile",
+    "ControlNet_v11p_sd15_lineart",
+    "AnimateDiff_v2",
+    "AnimateDiff_xl_beta",
+    "RIFE",
+    "BeautifulPrompt",
+    "opus-mt-zh-en",
+    "IP-Adapter-SD",
+    "IP-Adapter-SDXL",
+    "StableDiffusion3",
+    "StableDiffusion3_without_T5",
+    "Kolors",
+    "SDXL-vae-fp16-fix",
+    "ControlNet_union_sdxl_promax",
+    "FLUX.1-dev",
+    "FLUX.1-schnell",
+    "InstantX/FLUX.1-dev-Controlnet-Union-alpha",
+    "jasperai/Flux.1-dev-Controlnet-Depth",
+    "jasperai/Flux.1-dev-Controlnet-Surface-Normals",
+    "jasperai/Flux.1-dev-Controlnet-Upscaler",
+    "alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Alpha",
+    "alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Beta",
+    "Shakker-Labs/FLUX.1-dev-ControlNet-Depth",
+    "Shakker-Labs/FLUX.1-dev-ControlNet-Union-Pro",
+    "InstantX/FLUX.1-dev-IP-Adapter",
+    "SDXL_lora_zyd232_ChineseInkStyle_SDXL_v1_0",
+    "QwenPrompt",
+    "OmostPrompt",
+    "ESRGAN_x4",
+    "RIFE",
+    "OmniGen-v1",
+    "CogVideoX-5B",
+    "Annotators:Depth",
+    "Annotators:Softedge",
+    "Annotators:Lineart",
+    "Annotators:Normal",
+    "Annotators:Openpose",
+    "StableDiffusion3.5-large",
+    "StableDiffusion3.5-medium",
+    "HunyuanVideo",
+    "HunyuanVideo-fp8",
+]

diffsynth/controlnets/__init__.py

@@ -0,0 +1,2 @@
+from .controlnet_unit import ControlNetConfigUnit, ControlNetUnit, MultiControlNetManager, FluxMultiControlNetManager
+from .processors import Annotator

diffsynth/controlnets/controlnet_unit.py

@@ -0,0 +1,91 @@
+import torch
+import numpy as np
+from .processors import Processor_id
+
+
+class ControlNetConfigUnit:
+    def __init__(self, processor_id: Processor_id, model_path, scale=1.0, skip_processor=False):
+        self.processor_id = processor_id
+        self.model_path = model_path
+        self.scale = scale
+        self.skip_processor = skip_processor
+
+
+class ControlNetUnit:
+    def __init__(self, processor, model, scale=1.0):
+        self.processor = processor
+        self.model = model
+        self.scale = scale
+
+
+class MultiControlNetManager:
+    def __init__(self, controlnet_units=[]):
+        self.processors = [unit.processor for unit in controlnet_units]
+        self.models = [unit.model for unit in controlnet_units]
+        self.scales = [unit.scale for unit in controlnet_units]
+
+    def cpu(self):
+        for model in self.models:
+            model.cpu()
+
+    def to(self, device):
+        for model in self.models:
+            model.to(device)
+        for processor in self.processors:
+            processor.to(device)
+    
+    def process_image(self, image, processor_id=None):
+        if processor_id is None:
+            processed_image = [processor(image) for processor in self.processors]
+        else:
+            processed_image = [self.processors[processor_id](image)]
+        processed_image = torch.concat([
+            torch.Tensor(np.array(image_, dtype=np.float32) / 255).permute(2, 0, 1).unsqueeze(0)
+            for image_ in processed_image
+        ], dim=0)
+        return processed_image
+    
+    def __call__(
+        self,
+        sample, timestep, encoder_hidden_states, conditionings,
+        tiled=False, tile_size=64, tile_stride=32, **kwargs
+    ):
+        res_stack = None
+        for processor, conditioning, model, scale in zip(self.processors, conditionings, self.models, self.scales):
+            res_stack_ = model(
+                sample, timestep, encoder_hidden_states, conditioning, **kwargs,
+                tiled=tiled, tile_size=tile_size, tile_stride=tile_stride,
+                processor_id=processor.processor_id
+            )
+            res_stack_ = [res * scale for res in res_stack_]
+            if res_stack is None:
+                res_stack = res_stack_
+            else:
+                res_stack = [i + j for i, j in zip(res_stack, res_stack_)]
+        return res_stack
+
+
+class FluxMultiControlNetManager(MultiControlNetManager):
+    def __init__(self, controlnet_units=[]):
+        super().__init__(controlnet_units=controlnet_units)
+
+    def process_image(self, image, processor_id=None):
+        if processor_id is None:
+            processed_image = [processor(image) for processor in self.processors]
+        else:
+            processed_image = [self.processors[processor_id](image)]
+        return processed_image
+
+    def __call__(self, conditionings, **kwargs):
+        res_stack, single_res_stack = None, None
+        for processor, conditioning, model, scale in zip(self.processors, conditionings, self.models, self.scales):
+            res_stack_, single_res_stack_ = model(controlnet_conditioning=conditioning, processor_id=processor.processor_id, **kwargs)
+            res_stack_ = [res * scale for res in res_stack_]
+            single_res_stack_ = [res * scale for res in single_res_stack_]
+            if res_stack is None:
+                res_stack = res_stack_
+                single_res_stack = single_res_stack_
+            else:
+                res_stack = [i + j for i, j in zip(res_stack, res_stack_)]
+                single_res_stack = [i + j for i, j in zip(single_res_stack, single_res_stack_)]
+        return res_stack, single_res_stack

diffsynth/controlnets/processors.py

@@ -0,0 +1,62 @@
+from typing_extensions import Literal, TypeAlias
+
+
+Processor_id: TypeAlias = Literal[
+    "canny", "depth", "softedge", "lineart", "lineart_anime", "openpose", "normal", "tile", "none", "inpaint"
+]
+
+class Annotator:
+    def __init__(self, processor_id: Processor_id, model_path="models/Annotators", detect_resolution=None, device='cuda', skip_processor=False):
+        if not skip_processor:
+            if processor_id == "canny":
+                from controlnet_aux.processor import CannyDetector
+                self.processor = CannyDetector()
+            elif processor_id == "depth":
+                from controlnet_aux.processor import MidasDetector
+                self.processor = MidasDetector.from_pretrained(model_path).to(device)
+            elif processor_id == "softedge":
+                from controlnet_aux.processor import HEDdetector
+                self.processor = HEDdetector.from_pretrained(model_path).to(device)
+            elif processor_id == "lineart":
+                from controlnet_aux.processor import LineartDetector
+                self.processor = LineartDetector.from_pretrained(model_path).to(device)
+            elif processor_id == "lineart_anime":
+                from controlnet_aux.processor import LineartAnimeDetector
+                self.processor = LineartAnimeDetector.from_pretrained(model_path).to(device)
+            elif processor_id == "openpose":
+                from controlnet_aux.processor import OpenposeDetector
+                self.processor = OpenposeDetector.from_pretrained(model_path).to(device)
+            elif processor_id == "normal":
+                from controlnet_aux.processor import NormalBaeDetector
+                self.processor = NormalBaeDetector.from_pretrained(model_path).to(device)
+            elif processor_id == "tile" or processor_id == "none" or processor_id == "inpaint":
+                self.processor = None
+            else:
+                raise ValueError(f"Unsupported processor_id: {processor_id}")
+        else:
+            self.processor = None
+
+        self.processor_id = processor_id
+        self.detect_resolution = detect_resolution
+    
+    def to(self,device):
+        if hasattr(self.processor,"model") and hasattr(self.processor.model,"to"):
+
+            self.processor.model.to(device)
+
+    def __call__(self, image, mask=None):
+        width, height = image.size
+        if self.processor_id == "openpose":
+            kwargs = {
+                "include_body": True,
+                "include_hand": True,
+                "include_face": True
+            }
+        else:
+            kwargs = {}
+        if self.processor is not None:
+            detect_resolution = self.detect_resolution if self.detect_resolution is not None else min(width, height)
+            image = self.processor(image, detect_resolution=detect_resolution, image_resolution=min(width, height), **kwargs)
+        image = image.resize((width, height))
+        return image
+

diffsynth/data/__init__.py

@@ -0,0 +1 @@
+from .video import VideoData, save_video, save_frames

diffsynth/data/simple_text_image.py

@@ -0,0 +1,41 @@
+import torch, os, torchvision
+from torchvision import transforms
+import pandas as pd
+from PIL import Image
+
+
+
+class TextImageDataset(torch.utils.data.Dataset):
+    def __init__(self, dataset_path, steps_per_epoch=10000, height=1024, width=1024, center_crop=True, random_flip=False):
+        self.steps_per_epoch = steps_per_epoch
+        metadata = pd.read_csv(os.path.join(dataset_path, "train/metadata.csv"))
+        self.path = [os.path.join(dataset_path, "train", file_name) for file_name in metadata["file_name"]]
+        self.text = metadata["text"].to_list()
+        self.height = height
+        self.width = width
+        self.image_processor = transforms.Compose(
+            [
+                transforms.CenterCrop((height, width)) if center_crop else transforms.RandomCrop((height, width)),
+                transforms.RandomHorizontalFlip() if random_flip else transforms.Lambda(lambda x: x),
+                transforms.ToTensor(),
+                transforms.Normalize([0.5], [0.5]),
+            ]
+        )
+
+
+    def __getitem__(self, index):
+        data_id = torch.randint(0, len(self.path), (1,))[0]
+        data_id = (data_id + index) % len(self.path) # For fixed seed.
+        text = self.text[data_id]
+        image = Image.open(self.path[data_id]).convert("RGB")
+        target_height, target_width = self.height, self.width
+        width, height = image.size
+        scale = max(target_width / width, target_height / height)
+        shape = [round(height*scale),round(width*scale)]
+        image = torchvision.transforms.functional.resize(image,shape,interpolation=transforms.InterpolationMode.BILINEAR)
+        image = self.image_processor(image)
+        return {"text": text, "image": image}
+
+
+    def __len__(self):
+        return self.steps_per_epoch

diffsynth/data/video.py

@@ -0,0 +1,148 @@
+import imageio, os
+import numpy as np
+from PIL import Image
+from tqdm import tqdm
+
+
+class LowMemoryVideo:
+    def __init__(self, file_name):
+        self.reader = imageio.get_reader(file_name)
+    
+    def __len__(self):
+        return self.reader.count_frames()
+
+    def __getitem__(self, item):
+        return Image.fromarray(np.array(self.reader.get_data(item))).convert("RGB")
+
+    def __del__(self):
+        self.reader.close()
+
+
+def split_file_name(file_name):
+    result = []
+    number = -1
+    for i in file_name:
+        if ord(i)>=ord("0") and ord(i)<=ord("9"):
+            if number == -1:
+                number = 0
+            number = number*10 + ord(i) - ord("0")
+        else:
+            if number != -1:
+                result.append(number)
+                number = -1
+            result.append(i)
+    if number != -1:
+        result.append(number)
+    result = tuple(result)
+    return result
+
+
+def search_for_images(folder):
+    file_list = [i for i in os.listdir(folder) if i.endswith(".jpg") or i.endswith(".png")]
+    file_list = [(split_file_name(file_name), file_name) for file_name in file_list]
+    file_list = [i[1] for i in sorted(file_list)]
+    file_list = [os.path.join(folder, i) for i in file_list]
+    return file_list
+
+
+class LowMemoryImageFolder:
+    def __init__(self, folder, file_list=None):
+        if file_list is None:
+            self.file_list = search_for_images(folder)
+        else:
+            self.file_list = [os.path.join(folder, file_name) for file_name in file_list]
+    
+    def __len__(self):
+        return len(self.file_list)
+
+    def __getitem__(self, item):
+        return Image.open(self.file_list[item]).convert("RGB")
+
+    def __del__(self):
+        pass
+
+
+def crop_and_resize(image, height, width):
+    image = np.array(image)
+    image_height, image_width, _ = image.shape
+    if image_height / image_width < height / width:
+        croped_width = int(image_height / height * width)
+        left = (image_width - croped_width) // 2
+        image = image[:, left: left+croped_width]
+        image = Image.fromarray(image).resize((width, height))
+    else:
+        croped_height = int(image_width / width * height)
+        left = (image_height - croped_height) // 2
+        image = image[left: left+croped_height, :]
+        image = Image.fromarray(image).resize((width, height))
+    return image
+
+
+class VideoData:
+    def __init__(self, video_file=None, image_folder=None, height=None, width=None, **kwargs):
+        if video_file is not None:
+            self.data_type = "video"
+            self.data = LowMemoryVideo(video_file, **kwargs)
+        elif image_folder is not None:
+            self.data_type = "images"
+            self.data = LowMemoryImageFolder(image_folder, **kwargs)
+        else:
+            raise ValueError("Cannot open video or image folder")
+        self.length = None
+        self.set_shape(height, width)
+
+    def raw_data(self):
+        frames = []
+        for i in range(self.__len__()):
+            frames.append(self.__getitem__(i))
+        return frames
+
+    def set_length(self, length):
+        self.length = length
+
+    def set_shape(self, height, width):
+        self.height = height
+        self.width = width
+
+    def __len__(self):
+        if self.length is None:
+            return len(self.data)
+        else:
+            return self.length
+
+    def shape(self):
+        if self.height is not None and self.width is not None:
+            return self.height, self.width
+        else:
+            height, width, _ = self.__getitem__(0).shape
+            return height, width
+
+    def __getitem__(self, item):
+        frame = self.data.__getitem__(item)
+        width, height = frame.size
+        if self.height is not None and self.width is not None:
+            if self.height != height or self.width != width:
+                frame = crop_and_resize(frame, self.height, self.width)
+        return frame
+
+    def __del__(self):
+        pass
+
+    def save_images(self, folder):
+        os.makedirs(folder, exist_ok=True)
+        for i in tqdm(range(self.__len__()), desc="Saving images"):
+            frame = self.__getitem__(i)
+            frame.save(os.path.join(folder, f"{i}.png"))
+
+
+def save_video(frames, save_path, fps, quality=9, ffmpeg_params=None):
+    writer = imageio.get_writer(save_path, fps=fps, quality=quality, ffmpeg_params=ffmpeg_params)
+    for frame in tqdm(frames, desc="Saving video"):
+        frame = np.array(frame)
+        writer.append_data(frame)
+    writer.close()
+
+def save_frames(frames, save_path):
+    os.makedirs(save_path, exist_ok=True)
+    for i, frame in enumerate(tqdm(frames, desc="Saving images")):
+        frame.save(os.path.join(save_path, f"{i}.png"))

diffsynth/extensions/ESRGAN/__init__.py

@@ -0,0 +1,137 @@
+import torch
+from einops import repeat
+from PIL import Image
+import numpy as np
+
+
+class ResidualDenseBlock(torch.nn.Module):
+
+    def __init__(self, num_feat=64, num_grow_ch=32):
+        super(ResidualDenseBlock, self).__init__()
+        self.conv1 = torch.nn.Conv2d(num_feat, num_grow_ch, 3, 1, 1)
+        self.conv2 = torch.nn.Conv2d(num_feat + num_grow_ch, num_grow_ch, 3, 1, 1)
+        self.conv3 = torch.nn.Conv2d(num_feat + 2 * num_grow_ch, num_grow_ch, 3, 1, 1)
+        self.conv4 = torch.nn.Conv2d(num_feat + 3 * num_grow_ch, num_grow_ch, 3, 1, 1)
+        self.conv5 = torch.nn.Conv2d(num_feat + 4 * num_grow_ch, num_feat, 3, 1, 1)
+        self.lrelu = torch.nn.LeakyReLU(negative_slope=0.2, inplace=True)
+
+    def forward(self, x):
+        x1 = self.lrelu(self.conv1(x))
+        x2 = self.lrelu(self.conv2(torch.cat((x, x1), 1)))
+        x3 = self.lrelu(self.conv3(torch.cat((x, x1, x2), 1)))
+        x4 = self.lrelu(self.conv4(torch.cat((x, x1, x2, x3), 1)))
+        x5 = self.conv5(torch.cat((x, x1, x2, x3, x4), 1))
+        return x5 * 0.2 + x
+
+
+class RRDB(torch.nn.Module):
+
+    def __init__(self, num_feat, num_grow_ch=32):
+        super(RRDB, self).__init__()
+        self.rdb1 = ResidualDenseBlock(num_feat, num_grow_ch)
+        self.rdb2 = ResidualDenseBlock(num_feat, num_grow_ch)
+        self.rdb3 = ResidualDenseBlock(num_feat, num_grow_ch)
+
+    def forward(self, x):
+        out = self.rdb1(x)
+        out = self.rdb2(out)
+        out = self.rdb3(out)
+        return out * 0.2 + x
+
+
+class RRDBNet(torch.nn.Module):
+
+    def __init__(self, num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, **kwargs):
+        super(RRDBNet, self).__init__()
+        self.conv_first = torch.nn.Conv2d(num_in_ch, num_feat, 3, 1, 1)
+        self.body = torch.torch.nn.Sequential(*[RRDB(num_feat=num_feat, num_grow_ch=num_grow_ch) for _ in range(num_block)])
+        self.conv_body = torch.nn.Conv2d(num_feat, num_feat, 3, 1, 1)
+        # upsample
+        self.conv_up1 = torch.nn.Conv2d(num_feat, num_feat, 3, 1, 1)
+        self.conv_up2 = torch.nn.Conv2d(num_feat, num_feat, 3, 1, 1)
+        self.conv_hr = torch.nn.Conv2d(num_feat, num_feat, 3, 1, 1)
+        self.conv_last = torch.nn.Conv2d(num_feat, num_out_ch, 3, 1, 1)
+        self.lrelu = torch.nn.LeakyReLU(negative_slope=0.2, inplace=True)
+
+    def forward(self, x):
+        feat = x
+        feat = self.conv_first(feat)
+        body_feat = self.conv_body(self.body(feat))
+        feat = feat + body_feat
+        # upsample
+        feat = repeat(feat, "B C H W -> B C (H 2) (W 2)")
+        feat = self.lrelu(self.conv_up1(feat))
+        feat = repeat(feat, "B C H W -> B C (H 2) (W 2)")
+        feat = self.lrelu(self.conv_up2(feat))
+        out = self.conv_last(self.lrelu(self.conv_hr(feat)))
+        return out
+    
+    @staticmethod
+    def state_dict_converter():
+        return RRDBNetStateDictConverter()
+    
+
+class RRDBNetStateDictConverter:
+    def __init__(self):
+        pass
+
+    def from_diffusers(self, state_dict):
+        return state_dict, {"upcast_to_float32": True}
+    
+    def from_civitai(self, state_dict):
+        return state_dict, {"upcast_to_float32": True}
+
+
+class ESRGAN(torch.nn.Module):
+    def __init__(self, model):
+        super().__init__()
+        self.model = model
+
+    @staticmethod
+    def from_model_manager(model_manager):
+        return ESRGAN(model_manager.fetch_model("esrgan"))
+
+    def process_image(self, image):
+        image = torch.Tensor(np.array(image, dtype=np.float32) / 255).permute(2, 0, 1)
+        return image
+    
+    def process_images(self, images):
+        images = [self.process_image(image) for image in images]
+        images = torch.stack(images)
+        return images
+    
+    def decode_images(self, images):
+        images = (images.permute(0, 2, 3, 1) * 255).clip(0, 255).numpy().astype(np.uint8)
+        images = [Image.fromarray(image) for image in images]
+        return images
+    
+    @torch.no_grad()
+    def upscale(self, images, batch_size=4, progress_bar=lambda x:x):
+        if not isinstance(images, list):
+            images = [images]
+            is_single_image = True
+        else:
+            is_single_image = False
+
+        # Preprocess
+        input_tensor = self.process_images(images)
+
+        # Interpolate
+        output_tensor = []
+        for batch_id in progress_bar(range(0, input_tensor.shape[0], batch_size)):
+            batch_id_ = min(batch_id + batch_size, input_tensor.shape[0])
+            batch_input_tensor = input_tensor[batch_id: batch_id_]
+            batch_input_tensor = batch_input_tensor.to(
+                device=self.model.conv_first.weight.device,
+                dtype=self.model.conv_first.weight.dtype)
+            batch_output_tensor = self.model(batch_input_tensor)
+            output_tensor.append(batch_output_tensor.cpu())
+        
+        # Output
+        output_tensor = torch.concat(output_tensor, dim=0)
+
+        # To images
+        output_images = self.decode_images(output_tensor)
+        if is_single_image:
+            output_images = output_images[0]
+        return output_images

diffsynth/extensions/FastBlend/__init__.py

@@ -0,0 +1,63 @@
+from .runners.fast import TableManager, PyramidPatchMatcher
+from PIL import Image
+import numpy as np
+import cupy as cp
+
+
+class FastBlendSmoother:
+    def __init__(self):
+        self.batch_size = 8
+        self.window_size = 64
+        self.ebsynth_config = {
+            "minimum_patch_size": 5,
+            "threads_per_block": 8,
+            "num_iter": 5,
+            "gpu_id": 0,
+            "guide_weight": 10.0,
+            "initialize": "identity",
+            "tracking_window_size": 0,
+        }
+
+    @staticmethod
+    def from_model_manager(model_manager):
+        # TODO: fetch GPU ID from model_manager
+        return FastBlendSmoother()
+
+    def run(self, frames_guide, frames_style, batch_size, window_size, ebsynth_config):
+        frames_guide = [np.array(frame) for frame in frames_guide]
+        frames_style = [np.array(frame) for frame in frames_style]
+        table_manager = TableManager()
+        patch_match_engine = PyramidPatchMatcher(
+            image_height=frames_style[0].shape[0],
+            image_width=frames_style[0].shape[1],
+            channel=3,
+            **ebsynth_config
+        )
+        # left part
+        table_l = table_manager.build_remapping_table(frames_guide, frames_style, patch_match_engine, batch_size, desc="FastBlend Step 1/4")
+        table_l = table_manager.remapping_table_to_blending_table(table_l)
+        table_l = table_manager.process_window_sum(frames_guide, table_l, patch_match_engine, window_size, batch_size, desc="FastBlend Step 2/4")
+        # right part
+        table_r = table_manager.build_remapping_table(frames_guide[::-1], frames_style[::-1], patch_match_engine, batch_size, desc="FastBlend Step 3/4")
+        table_r = table_manager.remapping_table_to_blending_table(table_r)
+        table_r = table_manager.process_window_sum(frames_guide[::-1], table_r, patch_match_engine, window_size, batch_size, desc="FastBlend Step 4/4")[::-1]
+        # merge
+        frames = []
+        for (frame_l, weight_l), frame_m, (frame_r, weight_r) in zip(table_l, frames_style, table_r):
+            weight_m = -1
+            weight = weight_l + weight_m + weight_r
+            frame = frame_l * (weight_l / weight) + frame_m * (weight_m / weight) + frame_r * (weight_r / weight)
+            frames.append(frame)
+        frames = [Image.fromarray(frame.clip(0, 255).astype("uint8")) for frame in frames]
+        return frames
+    
+    def __call__(self, rendered_frames, original_frames=None, **kwargs):
+        frames = self.run(
+            original_frames, rendered_frames,
+            self.batch_size, self.window_size, self.ebsynth_config
+        )
+        mempool = cp.get_default_memory_pool()
+        pinned_mempool = cp.get_default_pinned_memory_pool()
+        mempool.free_all_blocks()
+        pinned_mempool.free_all_blocks()
+        return frames

diffsynth/extensions/FastBlend/api.py

@@ -0,0 +1,397 @@
+from .runners import AccurateModeRunner, FastModeRunner, BalancedModeRunner, InterpolationModeRunner, InterpolationModeSingleFrameRunner
+from .data import VideoData, get_video_fps, save_video, search_for_images
+import os
+import gradio as gr
+
+
+def check_input_for_blending(video_guide, video_guide_folder, video_style, video_style_folder):
+    frames_guide = VideoData(video_guide, video_guide_folder)
+    frames_style = VideoData(video_style, video_style_folder)
+    message = ""
+    if len(frames_guide) < len(frames_style):
+        message += f"The number of frames mismatches. Only the first {len(frames_guide)} frames of style video will be used.\n"
+        frames_style.set_length(len(frames_guide))
+    elif len(frames_guide) > len(frames_style):
+        message += f"The number of frames mismatches. Only the first {len(frames_style)} frames of guide video will be used.\n"
+        frames_guide.set_length(len(frames_style))
+    height_guide, width_guide = frames_guide.shape()
+    height_style, width_style = frames_style.shape()
+    if height_guide != height_style or width_guide != width_style:
+        message += f"The shape of frames mismatches. The frames in style video will be resized to (height: {height_guide}, width: {width_guide})\n"
+        frames_style.set_shape(height_guide, width_guide)
+    return frames_guide, frames_style, message
+
+
+def smooth_video(
+    video_guide,
+    video_guide_folder,
+    video_style,
+    video_style_folder,
+    mode,
+    window_size,
+    batch_size,
+    tracking_window_size,
+    output_path,
+    fps,
+    minimum_patch_size,
+    num_iter,
+    guide_weight,
+    initialize,
+    progress = None,
+):
+    # input
+    frames_guide, frames_style, message = check_input_for_blending(video_guide, video_guide_folder, video_style, video_style_folder)
+    if len(message) > 0:
+        print(message)
+    # output
+    if output_path == "":
+        if video_style is None:
+            output_path = os.path.join(video_style_folder, "output")
+        else:
+            output_path = os.path.join(os.path.split(video_style)[0], "output")
+        os.makedirs(output_path, exist_ok=True)
+        print("No valid output_path. Your video will be saved here:", output_path)
+    elif not os.path.exists(output_path):
+        os.makedirs(output_path, exist_ok=True)
+        print("Your video will be saved here:", output_path)
+    frames_path = os.path.join(output_path, "frames")
+    video_path = os.path.join(output_path, "video.mp4")
+    os.makedirs(frames_path, exist_ok=True)
+    # process
+    if mode == "Fast" or mode == "Balanced":
+        tracking_window_size = 0
+    ebsynth_config = {
+        "minimum_patch_size": minimum_patch_size,
+        "threads_per_block": 8,
+        "num_iter": num_iter,
+        "gpu_id": 0,
+        "guide_weight": guide_weight,
+        "initialize": initialize,
+        "tracking_window_size": tracking_window_size,
+    }
+    if mode == "Fast":
+        FastModeRunner().run(frames_guide, frames_style, batch_size=batch_size, window_size=window_size, ebsynth_config=ebsynth_config, save_path=frames_path)
+    elif mode == "Balanced":
+        BalancedModeRunner().run(frames_guide, frames_style, batch_size=batch_size, window_size=window_size, ebsynth_config=ebsynth_config, save_path=frames_path)
+    elif mode == "Accurate":
+        AccurateModeRunner().run(frames_guide, frames_style, batch_size=batch_size, window_size=window_size, ebsynth_config=ebsynth_config, save_path=frames_path)
+    # output
+    try:
+        fps = int(fps)
+    except:
+        fps = get_video_fps(video_style) if video_style is not None else 30
+    print("Fps:", fps)
+    print("Saving video...")
+    video_path = save_video(frames_path, video_path, num_frames=len(frames_style), fps=fps)
+    print("Success!")
+    print("Your frames are here:", frames_path)
+    print("Your video is here:", video_path)
+    return output_path, fps, video_path
+
+
+class KeyFrameMatcher:
+    def __init__(self):
+        pass
+
+    def extract_number_from_filename(self, file_name):
+        result = []
+        number = -1
+        for i in file_name:
+            if ord(i)>=ord("0") and ord(i)<=ord("9"):
+                if number == -1:
+                    number = 0
+                number = number*10 + ord(i) - ord("0")
+            else:
+                if number != -1:
+                    result.append(number)
+                    number = -1
+        if number != -1:
+            result.append(number)
+        result = tuple(result)
+        return result
+
+    def extract_number_from_filenames(self, file_names):
+        numbers = [self.extract_number_from_filename(file_name) for file_name in file_names]
+        min_length = min(len(i) for i in numbers)
+        for i in range(min_length-1, -1, -1):
+            if len(set(number[i] for number in numbers))==len(file_names):
+                return [number[i] for number in numbers]
+        return list(range(len(file_names)))
+
+    def match_using_filename(self, file_names_a, file_names_b):
+        file_names_b_set = set(file_names_b)
+        matched_file_name = []
+        for file_name in file_names_a:
+            if file_name not in file_names_b_set:
+                matched_file_name.append(None)
+            else:
+                matched_file_name.append(file_name)
+        return matched_file_name
+
+    def match_using_numbers(self, file_names_a, file_names_b):
+        numbers_a = self.extract_number_from_filenames(file_names_a)
+        numbers_b = self.extract_number_from_filenames(file_names_b)
+        numbers_b_dict = {number: file_name for number, file_name in zip(numbers_b, file_names_b)}
+        matched_file_name = []
+        for number in numbers_a:
+            if number in numbers_b_dict:
+                matched_file_name.append(numbers_b_dict[number])
+            else:
+                matched_file_name.append(None)
+        return matched_file_name
+
+    def match_filenames(self, file_names_a, file_names_b):
+        matched_file_name = self.match_using_filename(file_names_a, file_names_b)
+        if sum([i is not None for i in matched_file_name]) > 0:
+            return matched_file_name
+        matched_file_name = self.match_using_numbers(file_names_a, file_names_b)
+        return matched_file_name
+
+
+def detect_frames(frames_path, keyframes_path):
+    if not os.path.exists(frames_path) and not os.path.exists(keyframes_path):
+        return "Please input the directory of guide video and rendered frames"
+    elif not os.path.exists(frames_path):
+        return "Please input the directory of guide video"
+    elif not os.path.exists(keyframes_path):
+        return "Please input the directory of rendered frames"
+    frames = [os.path.split(i)[-1] for i in search_for_images(frames_path)]
+    keyframes = [os.path.split(i)[-1] for i in search_for_images(keyframes_path)]
+    if len(frames)==0:
+        return f"No images detected in {frames_path}"
+    if len(keyframes)==0:
+        return f"No images detected in {keyframes_path}"
+    matched_keyframes = KeyFrameMatcher().match_filenames(frames, keyframes)
+    max_filename_length = max([len(i) for i in frames])
+    if sum([i is not None for i in matched_keyframes])==0:
+        message = ""
+        for frame, matched_keyframe in zip(frames, matched_keyframes):
+            message += frame + " " * (max_filename_length - len(frame) + 1)
+            message += "--> No matched keyframes\n"
+    else:
+        message = ""
+        for frame, matched_keyframe in zip(frames, matched_keyframes):
+            message += frame + " " * (max_filename_length - len(frame) + 1)
+            if matched_keyframe is None:
+                message += "--> [to be rendered]\n"
+            else:
+                message += f"--> {matched_keyframe}\n"
+    return message
+
+
+def check_input_for_interpolating(frames_path, keyframes_path):
+    # search for images
+    frames = [os.path.split(i)[-1] for i in search_for_images(frames_path)]
+    keyframes = [os.path.split(i)[-1] for i in search_for_images(keyframes_path)]
+    # match frames
+    matched_keyframes = KeyFrameMatcher().match_filenames(frames, keyframes)
+    file_list = [file_name for file_name in matched_keyframes if file_name is not None]
+    index_style = [i for i, file_name in enumerate(matched_keyframes) if file_name is not None]
+    frames_guide = VideoData(None, frames_path)
+    frames_style = VideoData(None, keyframes_path, file_list=file_list)
+    # match shape
+    message = ""
+    height_guide, width_guide = frames_guide.shape()
+    height_style, width_style = frames_style.shape()
+    if height_guide != height_style or width_guide != width_style:
+        message += f"The shape of frames mismatches. The rendered keyframes will be resized to (height: {height_guide}, width: {width_guide})\n"
+        frames_style.set_shape(height_guide, width_guide)
+    return frames_guide, frames_style, index_style, message
+
+
+def interpolate_video(
+    frames_path,
+    keyframes_path,
+    output_path,
+    fps,
+    batch_size,
+    tracking_window_size,
+    minimum_patch_size,
+    num_iter,
+    guide_weight,
+    initialize,
+    progress = None,
+):
+    # input
+    frames_guide, frames_style, index_style, message = check_input_for_interpolating(frames_path, keyframes_path)
+    if len(message) > 0:
+        print(message)
+    # output
+    if output_path == "":
+        output_path = os.path.join(keyframes_path, "output")
+        os.makedirs(output_path, exist_ok=True)
+        print("No valid output_path. Your video will be saved here:", output_path)
+    elif not os.path.exists(output_path):
+        os.makedirs(output_path, exist_ok=True)
+        print("Your video will be saved here:", output_path)
+    output_frames_path = os.path.join(output_path, "frames")
+    output_video_path = os.path.join(output_path, "video.mp4")
+    os.makedirs(output_frames_path, exist_ok=True)
+    # process
+    ebsynth_config = {
+        "minimum_patch_size": minimum_patch_size,
+        "threads_per_block": 8,
+        "num_iter": num_iter,
+        "gpu_id": 0,
+        "guide_weight": guide_weight,
+        "initialize": initialize,
+        "tracking_window_size": tracking_window_size
+    }
+    if len(index_style)==1:
+        InterpolationModeSingleFrameRunner().run(frames_guide, frames_style, index_style, batch_size=batch_size, ebsynth_config=ebsynth_config, save_path=output_frames_path)
+    else:
+        InterpolationModeRunner().run(frames_guide, frames_style, index_style, batch_size=batch_size, ebsynth_config=ebsynth_config, save_path=output_frames_path)
+    try:
+        fps = int(fps)
+    except:
+        fps = 30
+    print("Fps:", fps)
+    print("Saving video...")
+    video_path = save_video(output_frames_path, output_video_path, num_frames=len(frames_guide), fps=fps)
+    print("Success!")
+    print("Your frames are here:", output_frames_path)
+    print("Your video is here:", video_path)
+    return output_path, fps, video_path
+
+
+def on_ui_tabs():
+    with gr.Blocks(analytics_enabled=False) as ui_component:
+        with gr.Tab("Blend"):
+            gr.Markdown("""
+# Blend
+
+Given a guide video and a style video, this algorithm will make the style video fluent according to the motion features of the guide video. Click [here](https://github.com/Artiprocher/sd-webui-fastblend/assets/35051019/208d902d-6aba-48d7-b7d5-cd120ebd306d) to see the example. Note that this extension doesn't support long videos. Please use short videos (e.g., several seconds). The algorithm is mainly designed for 512*512 resolution. Please use a larger `Minimum patch size` for higher resolution.
+            """)
+            with gr.Row():
+                with gr.Column():
+                    with gr.Tab("Guide video"):
+                        video_guide = gr.Video(label="Guide video")
+                    with gr.Tab("Guide video (images format)"):
+                        video_guide_folder = gr.Textbox(label="Guide video (images format)", value="")
+                with gr.Column():
+                    with gr.Tab("Style video"):
+                        video_style = gr.Video(label="Style video")
+                    with gr.Tab("Style video (images format)"):
+                        video_style_folder = gr.Textbox(label="Style video (images format)", value="")
+                with gr.Column():
+                    output_path = gr.Textbox(label="Output directory", value="", placeholder="Leave empty to use the directory of style video")
+                    fps = gr.Textbox(label="Fps", value="", placeholder="Leave empty to use the default fps")
+                    video_output = gr.Video(label="Output video", interactive=False, show_share_button=True)
+            btn = gr.Button(value="Blend")
+            with gr.Row():
+                with gr.Column():
+                    gr.Markdown("# Settings")
+                    mode = gr.Radio(["Fast", "Balanced", "Accurate"], label="Inference mode", value="Fast", interactive=True)
+                    window_size = gr.Slider(label="Sliding window size", value=15, minimum=1, maximum=1000, step=1, interactive=True)
+                    batch_size = gr.Slider(label="Batch size", value=8, minimum=1, maximum=128, step=1, interactive=True)
+                    tracking_window_size = gr.Slider(label="Tracking window size (only for accurate mode)", value=0, minimum=0, maximum=10, step=1, interactive=True)
+                    gr.Markdown("## Advanced Settings")
+                    minimum_patch_size = gr.Slider(label="Minimum patch size (odd number)", value=5, minimum=5, maximum=99, step=2, interactive=True)
+                    num_iter = gr.Slider(label="Number of iterations", value=5, minimum=1, maximum=10, step=1, interactive=True)
+                    guide_weight = gr.Slider(label="Guide weight", value=10.0, minimum=0.0, maximum=100.0, step=0.1, interactive=True)
+                    initialize = gr.Radio(["identity", "random"], label="NNF initialization", value="identity", interactive=True)
+                with gr.Column():
+                    gr.Markdown("""
+# Reference
+
+* Output directory: the directory to save the video.
+* Inference mode
+
+|Mode|Time|Memory|Quality|Frame by frame output|Description|
+|-|-|-|-|-|-|
+|Fast|■|■■■|■■|No|Blend the frames using a tree-like data structure, which requires much RAM but is fast.|
+|Balanced|■■|■|■■|Yes|Blend the frames naively.|
+|Accurate|■■■|■|■■■|Yes|Blend the frames and align them together for higher video quality. When [batch size] >= [sliding window size] * 2 + 1, the performance is the best.|
+
+* Sliding window size: our algorithm will blend the frames in a sliding windows. If the size is n, each frame will be blended with the last n frames and the next n frames. A large sliding window can make the video fluent but sometimes smoggy.
+* Batch size: a larger batch size makes the program faster but requires more VRAM.
+* Tracking window size (only for accurate mode): The size of window in which our algorithm tracks moving objects. Empirically, 1 is enough.
+* Advanced settings
+    * Minimum patch size (odd number): the minimum patch size used for patch matching. (Default: 5)
+    * Number of iterations: the number of iterations of patch matching. (Default: 5)
+    * Guide weight: a parameter that determines how much motion feature applied to the style video. (Default: 10)
+    * NNF initialization: how to initialize the NNF (Nearest Neighbor Field). (Default: identity)
+                    """)
+            btn.click(
+                smooth_video,
+                inputs=[
+                    video_guide,
+                    video_guide_folder,
+                    video_style,
+                    video_style_folder,
+                    mode,
+                    window_size,
+                    batch_size,
+                    tracking_window_size,
+                    output_path,
+                    fps,
+                    minimum_patch_size,
+                    num_iter,
+                    guide_weight,
+                    initialize
+                ],
+                outputs=[output_path, fps, video_output]
+            )
+        with gr.Tab("Interpolate"):
+            gr.Markdown("""
+# Interpolate
+
+Given a guide video and some rendered keyframes, this algorithm will render the remaining frames. Click [here](https://github.com/Artiprocher/sd-webui-fastblend/assets/35051019/3490c5b4-8f67-478f-86de-f9adc2ace16a) to see the example. The algorithm is experimental and is only tested for 512*512 resolution.
+            """)
+            with gr.Row():
+                with gr.Column():
+                    with gr.Row():
+                        with gr.Column():
+                            video_guide_folder_ = gr.Textbox(label="Guide video (images format)", value="")
+                        with gr.Column():
+                            rendered_keyframes_ = gr.Textbox(label="Rendered keyframes (images format)", value="")
+                    with gr.Row():
+                        detected_frames = gr.Textbox(label="Detected frames", value="Please input the directory of guide video and rendered frames", lines=9, max_lines=9, interactive=False)
+                    video_guide_folder_.change(detect_frames, inputs=[video_guide_folder_, rendered_keyframes_], outputs=detected_frames)
+                    rendered_keyframes_.change(detect_frames, inputs=[video_guide_folder_, rendered_keyframes_], outputs=detected_frames)
+                with gr.Column():
+                    output_path_ = gr.Textbox(label="Output directory", value="", placeholder="Leave empty to use the directory of rendered keyframes")
+                    fps_ = gr.Textbox(label="Fps", value="", placeholder="Leave empty to use the default fps")
+                    video_output_ = gr.Video(label="Output video", interactive=False, show_share_button=True)
+            btn_ = gr.Button(value="Interpolate")
+            with gr.Row():
+                with gr.Column():
+                    gr.Markdown("# Settings")
+                    batch_size_ = gr.Slider(label="Batch size", value=8, minimum=1, maximum=128, step=1, interactive=True)
+                    tracking_window_size_ = gr.Slider(label="Tracking window size", value=0, minimum=0, maximum=10, step=1, interactive=True)
+                    gr.Markdown("## Advanced Settings")
+                    minimum_patch_size_ = gr.Slider(label="Minimum patch size (odd number, larger is better)", value=15, minimum=5, maximum=99, step=2, interactive=True)
+                    num_iter_ = gr.Slider(label="Number of iterations", value=5, minimum=1, maximum=10, step=1, interactive=True)
+                    guide_weight_ = gr.Slider(label="Guide weight", value=10.0, minimum=0.0, maximum=100.0, step=0.1, interactive=True)
+                    initialize_ = gr.Radio(["identity", "random"], label="NNF initialization", value="identity", interactive=True)
+                with gr.Column():
+                    gr.Markdown("""
+# Reference
+
+* Output directory: the directory to save the video.
+* Batch size: a larger batch size makes the program faster but requires more VRAM.
+* Tracking window size (only for accurate mode): The size of window in which our algorithm tracks moving objects. Empirically, 1 is enough.
+* Advanced settings
+    * Minimum patch size (odd number): the minimum patch size used for patch matching. **This parameter should be larger than that in blending. (Default: 15)**
+    * Number of iterations: the number of iterations of patch matching. (Default: 5)
+    * Guide weight: a parameter that determines how much motion feature applied to the style video. (Default: 10)
+    * NNF initialization: how to initialize the NNF (Nearest Neighbor Field). (Default: identity)
+                    """)
+            btn_.click(
+                interpolate_video,
+                inputs=[
+                    video_guide_folder_,
+                    rendered_keyframes_,
+                    output_path_,
+                    fps_,
+                    batch_size_,
+                    tracking_window_size_,
+                    minimum_patch_size_,
+                    num_iter_,
+                    guide_weight_,
+                    initialize_,
+                ],
+                outputs=[output_path_, fps_, video_output_]
+            )
+
+        return [(ui_component, "FastBlend", "FastBlend_ui")]

diffsynth/extensions/FastBlend/cupy_kernels.py

@@ -0,0 +1,119 @@
+import cupy as cp
+
+remapping_kernel = cp.RawKernel(r'''
+extern "C" __global__
+void remap(
+    const int height,
+    const int width,
+    const int channel,
+    const int patch_size,
+    const int pad_size,
+    const float* source_style,
+    const int* nnf,
+    float* target_style
+) {
+    const int r = (patch_size - 1) / 2;
+    const int x = blockDim.x * blockIdx.x + threadIdx.x;
+    const int y = blockDim.y * blockIdx.y + threadIdx.y;
+    if (x >= height or y >= width) return;
+    const int z = blockIdx.z * (height + pad_size * 2) * (width + pad_size * 2) * channel;
+    const int pid = (x + pad_size) * (width + pad_size * 2) + (y + pad_size);
+    const int min_px = x < r ? -x : -r;
+    const int max_px = x + r > height - 1 ? height - 1 - x : r;
+    const int min_py = y < r ? -y : -r;
+    const int max_py = y + r > width - 1 ? width - 1 - y : r;
+    int num = 0;
+    for (int px = min_px; px <= max_px; px++){
+        for (int py = min_py; py <= max_py; py++){
+            const int nid = (x + px) * width + y + py;
+            const int x_ = nnf[blockIdx.z * height * width * 2 + nid*2 + 0] - px;
+            const int y_ = nnf[blockIdx.z * height * width * 2 + nid*2 + 1] - py;
+            if (x_ < 0 or y_ < 0 or x_ >= height or y_ >= width)continue;
+            const int pid_ = (x_ + pad_size) * (width + pad_size * 2) + (y_ + pad_size);
+            num++;
+            for (int c = 0; c < channel; c++){
+                target_style[z + pid * channel + c] += source_style[z + pid_ * channel + c];
+            }
+        }
+    }
+    for (int c = 0; c < channel; c++){
+        target_style[z + pid * channel + c] /= num;
+    }
+}
+''', 'remap')
+
+
+patch_error_kernel = cp.RawKernel(r'''
+extern "C" __global__
+void patch_error(
+    const int height,
+    const int width,
+    const int channel,
+    const int patch_size,
+    const int pad_size,
+    const float* source,
+    const int* nnf,
+    const float* target,
+    float* error
+) {
+    const int r = (patch_size - 1) / 2;
+    const int x = blockDim.x * blockIdx.x + threadIdx.x;
+    const int y = blockDim.y * blockIdx.y + threadIdx.y;
+    const int z = blockIdx.z * (height + pad_size * 2) * (width + pad_size * 2) * channel;
+    if (x >= height or y >= width) return;
+    const int x_ = nnf[blockIdx.z * height * width * 2 + (x * width + y)*2 + 0];
+    const int y_ = nnf[blockIdx.z * height * width * 2 + (x * width + y)*2 + 1];
+    float e = 0;
+    for (int px = -r; px <= r; px++){
+        for (int py = -r; py <= r; py++){
+            const int pid = (x + pad_size + px) * (width + pad_size * 2) + y + pad_size + py;
+            const int pid_ = (x_ + pad_size + px) * (width + pad_size * 2) + y_ + pad_size + py;
+            for (int c = 0; c < channel; c++){
+                const float diff = target[z + pid * channel + c] - source[z + pid_ * channel + c];
+                e += diff * diff;
+            }
+        }
+    }
+    error[blockIdx.z * height * width + x * width + y] = e;
+}
+''', 'patch_error')
+
+
+pairwise_patch_error_kernel = cp.RawKernel(r'''
+extern "C" __global__
+void pairwise_patch_error(
+    const int height,
+    const int width,
+    const int channel,
+    const int patch_size,
+    const int pad_size,
+    const float* source_a,
+    const int* nnf_a,
+    const float* source_b,
+    const int* nnf_b,
+    float* error
+) {
+    const int r = (patch_size - 1) / 2;
+    const int x = blockDim.x * blockIdx.x + threadIdx.x;
+    const int y = blockDim.y * blockIdx.y + threadIdx.y;
+    const int z = blockIdx.z * (height + pad_size * 2) * (width + pad_size * 2) * channel;
+    if (x >= height or y >= width) return;
+    const int z_nnf = blockIdx.z * height * width * 2 + (x * width + y) * 2;
+    const int x_a = nnf_a[z_nnf + 0];
+    const int y_a = nnf_a[z_nnf + 1];
+    const int x_b = nnf_b[z_nnf + 0];
+    const int y_b = nnf_b[z_nnf + 1];
+    float e = 0;
+    for (int px = -r; px <= r; px++){
+        for (int py = -r; py <= r; py++){
+            const int pid_a = (x_a + pad_size + px) * (width + pad_size * 2) + y_a + pad_size + py;
+            const int pid_b = (x_b + pad_size + px) * (width + pad_size * 2) + y_b + pad_size + py;
+            for (int c = 0; c < channel; c++){
+                const float diff = source_a[z + pid_a * channel + c] - source_b[z + pid_b * channel + c];
+                e += diff * diff;
+            }
+        }
+    }
+    error[blockIdx.z * height * width + x * width + y] = e;
+}
+''', 'pairwise_patch_error')

diffsynth/extensions/FastBlend/data.py

@@ -0,0 +1,146 @@
+import imageio, os
+import numpy as np
+from PIL import Image
+
+
+def read_video(file_name):
+    reader = imageio.get_reader(file_name)
+    video = []
+    for frame in reader:
+        frame = np.array(frame)
+        video.append(frame)
+    reader.close()
+    return video
+
+
+def get_video_fps(file_name):
+    reader = imageio.get_reader(file_name)
+    fps = reader.get_meta_data()["fps"]
+    reader.close()
+    return fps
+
+
+def save_video(frames_path, video_path, num_frames, fps):
+    writer = imageio.get_writer(video_path, fps=fps, quality=9)
+    for i in range(num_frames):
+        frame = np.array(Image.open(os.path.join(frames_path, "%05d.png" % i)))
+        writer.append_data(frame)
+    writer.close()
+    return video_path
+
+
+class LowMemoryVideo:
+    def __init__(self, file_name):
+        self.reader = imageio.get_reader(file_name)
+    
+    def __len__(self):
+        return self.reader.count_frames()
+
+    def __getitem__(self, item):
+        return np.array(self.reader.get_data(item))
+
+    def __del__(self):
+        self.reader.close()
+
+
+def split_file_name(file_name):
+    result = []
+    number = -1
+    for i in file_name:
+        if ord(i)>=ord("0") and ord(i)<=ord("9"):
+            if number == -1:
+                number = 0
+            number = number*10 + ord(i) - ord("0")
+        else:
+            if number != -1:
+                result.append(number)
+                number = -1
+            result.append(i)
+    if number != -1:
+        result.append(number)
+    result = tuple(result)
+    return result
+
+
+def search_for_images(folder):
+    file_list = [i for i in os.listdir(folder) if i.endswith(".jpg") or i.endswith(".png")]
+    file_list = [(split_file_name(file_name), file_name) for file_name in file_list]
+    file_list = [i[1] for i in sorted(file_list)]
+    file_list = [os.path.join(folder, i) for i in file_list]
+    return file_list
+
+
+def read_images(folder):
+    file_list = search_for_images(folder)
+    frames = [np.array(Image.open(i)) for i in file_list]
+    return frames
+
+
+class LowMemoryImageFolder:
+    def __init__(self, folder, file_list=None):
+        if file_list is None:
+            self.file_list = search_for_images(folder)
+        else:
+            self.file_list = [os.path.join(folder, file_name) for file_name in file_list]
+    
+    def __len__(self):
+        return len(self.file_list)
+
+    def __getitem__(self, item):
+        return np.array(Image.open(self.file_list[item]))
+
+    def __del__(self):
+        pass
+
+
+class VideoData:
+    def __init__(self, video_file, image_folder, **kwargs):
+        if video_file is not None:
+            self.data_type = "video"
+            self.data = LowMemoryVideo(video_file, **kwargs)
+        elif image_folder is not None:
+            self.data_type = "images"
+            self.data = LowMemoryImageFolder(image_folder, **kwargs)
+        else:
+            raise ValueError("Cannot open video or image folder")
+        self.length = None
+        self.height = None
+        self.width = None
+
+    def raw_data(self):
+        frames = []
+        for i in range(self.__len__()):
+            frames.append(self.__getitem__(i))
+        return frames
+
+    def set_length(self, length):
+        self.length = length
+
+    def set_shape(self, height, width):
+        self.height = height
+        self.width = width
+
+    def __len__(self):
+        if self.length is None:
+            return len(self.data)
+        else:
+            return self.length
+
+    def shape(self):
+        if self.height is not None and self.width is not None:
+            return self.height, self.width
+        else:
+            height, width, _ = self.__getitem__(0).shape
+            return height, width
+
+    def __getitem__(self, item):
+        frame = self.data.__getitem__(item)
+        height, width, _ = frame.shape
+        if self.height is not None and self.width is not None:
+            if self.height != height or self.width != width:
+                frame = Image.fromarray(frame).resize((self.width, self.height))
+                frame = np.array(frame)
+        return frame
+
+    def __del__(self):
+        pass

diffsynth/extensions/FastBlend/patch_match.py

@@ -0,0 +1,298 @@
+from .cupy_kernels import remapping_kernel, patch_error_kernel, pairwise_patch_error_kernel
+import numpy as np
+import cupy as cp
+import cv2
+
+
+class PatchMatcher:
+    def __init__(
+        self, height, width, channel, minimum_patch_size,
+        threads_per_block=8, num_iter=5, gpu_id=0, guide_weight=10.0,
+        random_search_steps=3, random_search_range=4,
+        use_mean_target_style=False, use_pairwise_patch_error=False,
+        tracking_window_size=0
+    ):
+        self.height = height
+        self.width = width
+        self.channel = channel
+        self.minimum_patch_size = minimum_patch_size
+        self.threads_per_block = threads_per_block
+        self.num_iter = num_iter
+        self.gpu_id = gpu_id
+        self.guide_weight = guide_weight
+        self.random_search_steps = random_search_steps
+        self.random_search_range = random_search_range
+        self.use_mean_target_style = use_mean_target_style
+        self.use_pairwise_patch_error = use_pairwise_patch_error
+        self.tracking_window_size = tracking_window_size
+
+        self.patch_size_list = [minimum_patch_size + i*2 for i in range(num_iter)][::-1]
+        self.pad_size = self.patch_size_list[0] // 2
+        self.grid = (
+            (height + threads_per_block - 1) // threads_per_block,
+            (width + threads_per_block - 1) // threads_per_block
+        )
+        self.block = (threads_per_block, threads_per_block)
+
+    def pad_image(self, image):
+        return cp.pad(image, ((0, 0), (self.pad_size, self.pad_size), (self.pad_size, self.pad_size), (0, 0)))
+
+    def unpad_image(self, image):
+        return image[:, self.pad_size: -self.pad_size, self.pad_size: -self.pad_size, :]
+
+    def apply_nnf_to_image(self, nnf, source):
+        batch_size = source.shape[0]
+        target = cp.zeros((batch_size, self.height + self.pad_size * 2, self.width + self.pad_size * 2, self.channel), dtype=cp.float32)
+        remapping_kernel(
+            self.grid + (batch_size,),
+            self.block,
+            (self.height, self.width, self.channel, self.patch_size, self.pad_size, source, nnf, target)
+        )
+        return target
+
+    def get_patch_error(self, source, nnf, target):
+        batch_size = source.shape[0]
+        error = cp.zeros((batch_size, self.height, self.width), dtype=cp.float32)
+        patch_error_kernel(
+            self.grid + (batch_size,),
+            self.block,
+            (self.height, self.width, self.channel, self.patch_size, self.pad_size, source, nnf, target, error)
+        )
+        return error
+
+    def get_pairwise_patch_error(self, source, nnf):
+        batch_size = source.shape[0]//2
+        error = cp.zeros((batch_size, self.height, self.width), dtype=cp.float32)
+        source_a, nnf_a = source[0::2].copy(), nnf[0::2].copy()
+        source_b, nnf_b = source[1::2].copy(), nnf[1::2].copy()
+        pairwise_patch_error_kernel(
+            self.grid + (batch_size,),
+            self.block,
+            (self.height, self.width, self.channel, self.patch_size, self.pad_size, source_a, nnf_a, source_b, nnf_b, error)
+        )
+        error = error.repeat(2, axis=0)
+        return error
+
+    def get_error(self, source_guide, target_guide, source_style, target_style, nnf):
+        error_guide = self.get_patch_error(source_guide, nnf, target_guide)
+        if self.use_mean_target_style:
+            target_style = self.apply_nnf_to_image(nnf, source_style)
+            target_style = target_style.mean(axis=0, keepdims=True)
+            target_style = target_style.repeat(source_guide.shape[0], axis=0)
+        if self.use_pairwise_patch_error:
+            error_style = self.get_pairwise_patch_error(source_style, nnf)
+        else:
+            error_style = self.get_patch_error(source_style, nnf, target_style)
+        error = error_guide * self.guide_weight + error_style
+        return error
+
+    def clamp_bound(self, nnf):
+        nnf[:,:,:,0] = cp.clip(nnf[:,:,:,0], 0, self.height-1)
+        nnf[:,:,:,1] = cp.clip(nnf[:,:,:,1], 0, self.width-1)
+        return nnf
+
+    def random_step(self, nnf, r):
+        batch_size = nnf.shape[0]
+        step = cp.random.randint(-r, r+1, size=(batch_size, self.height, self.width, 2), dtype=cp.int32)
+        upd_nnf = self.clamp_bound(nnf + step)
+        return upd_nnf
+
+    def neighboor_step(self, nnf, d):
+        if d==0:
+            upd_nnf = cp.concatenate([nnf[:, :1, :], nnf[:, :-1, :]], axis=1)
+            upd_nnf[:, :, :, 0] += 1
+        elif d==1:
+            upd_nnf = cp.concatenate([nnf[:, :, :1], nnf[:, :, :-1]], axis=2)
+            upd_nnf[:, :, :, 1] += 1
+        elif d==2:
+            upd_nnf = cp.concatenate([nnf[:, 1:, :], nnf[:, -1:, :]], axis=1)
+            upd_nnf[:, :, :, 0] -= 1
+        elif d==3:
+            upd_nnf = cp.concatenate([nnf[:, :, 1:], nnf[:, :, -1:]], axis=2)
+            upd_nnf[:, :, :, 1] -= 1
+        upd_nnf = self.clamp_bound(upd_nnf)
+        return upd_nnf
+        
+    def shift_nnf(self, nnf, d):
+        if d>0:
+            d = min(nnf.shape[0], d)
+            upd_nnf = cp.concatenate([nnf[d:]] + [nnf[-1:]] * d, axis=0)
+        else:
+            d = max(-nnf.shape[0], d)
+            upd_nnf = cp.concatenate([nnf[:1]] * (-d) + [nnf[:d]], axis=0)
+        return upd_nnf
+    
+    def track_step(self, nnf, d):
+        if self.use_pairwise_patch_error:
+            upd_nnf = cp.zeros_like(nnf)
+            upd_nnf[0::2] = self.shift_nnf(nnf[0::2], d)
+            upd_nnf[1::2] = self.shift_nnf(nnf[1::2], d)
+        else:
+            upd_nnf = self.shift_nnf(nnf, d)
+        return upd_nnf
+
+    def C(self, n, m):
+        # not used
+        c = 1
+        for i in range(1, n+1):
+            c *= i
+        for i in range(1, m+1):
+            c //= i
+        for i in range(1, n-m+1):
+            c //= i
+        return c
+
+    def bezier_step(self, nnf, r):
+        # not used
+        n = r * 2 - 1
+        upd_nnf = cp.zeros(shape=nnf.shape, dtype=cp.float32)
+        for i, d in enumerate(list(range(-r, 0)) + list(range(1, r+1))):
+            if d>0:
+                ctl_nnf = cp.concatenate([nnf[d:]] + [nnf[-1:]] * d, axis=0)
+            elif d<0:
+                ctl_nnf = cp.concatenate([nnf[:1]] * (-d) + [nnf[:d]], axis=0)
+            upd_nnf += ctl_nnf * (self.C(n, i) / 2**n)
+        upd_nnf = self.clamp_bound(upd_nnf).astype(nnf.dtype)
+        return upd_nnf
+
+    def update(self, source_guide, target_guide, source_style, target_style, nnf, err, upd_nnf):
+        upd_err = self.get_error(source_guide, target_guide, source_style, target_style, upd_nnf)
+        upd_idx = (upd_err < err)
+        nnf[upd_idx] = upd_nnf[upd_idx]
+        err[upd_idx] = upd_err[upd_idx]
+        return nnf, err
+
+    def propagation(self, source_guide, target_guide, source_style, target_style, nnf, err):
+        for d in cp.random.permutation(4):
+            upd_nnf = self.neighboor_step(nnf, d)
+            nnf, err = self.update(source_guide, target_guide, source_style, target_style, nnf, err, upd_nnf)
+        return nnf, err
+        
+    def random_search(self, source_guide, target_guide, source_style, target_style, nnf, err):
+        for i in range(self.random_search_steps):
+            upd_nnf = self.random_step(nnf, self.random_search_range)
+            nnf, err = self.update(source_guide, target_guide, source_style, target_style, nnf, err, upd_nnf)
+        return nnf, err
+
+    def track(self, source_guide, target_guide, source_style, target_style, nnf, err):
+        for d in range(1, self.tracking_window_size + 1):
+            upd_nnf = self.track_step(nnf, d)
+            nnf, err = self.update(source_guide, target_guide, source_style, target_style, nnf, err, upd_nnf)
+            upd_nnf = self.track_step(nnf, -d)
+            nnf, err = self.update(source_guide, target_guide, source_style, target_style, nnf, err, upd_nnf)
+        return nnf, err
+
+    def iteration(self, source_guide, target_guide, source_style, target_style, nnf, err):
+        nnf, err = self.propagation(source_guide, target_guide, source_style, target_style, nnf, err)
+        nnf, err = self.random_search(source_guide, target_guide, source_style, target_style, nnf, err)
+        nnf, err = self.track(source_guide, target_guide, source_style, target_style, nnf, err)
+        return nnf, err
+
+    def estimate_nnf(self, source_guide, target_guide, source_style, nnf):
+        with cp.cuda.Device(self.gpu_id):
+            source_guide = self.pad_image(source_guide)
+            target_guide = self.pad_image(target_guide)
+            source_style = self.pad_image(source_style)
+            for it in range(self.num_iter):
+                self.patch_size = self.patch_size_list[it]
+                target_style = self.apply_nnf_to_image(nnf, source_style)
+                err = self.get_error(source_guide, target_guide, source_style, target_style, nnf)
+                nnf, err = self.iteration(source_guide, target_guide, source_style, target_style, nnf, err)
+            target_style = self.unpad_image(self.apply_nnf_to_image(nnf, source_style))
+        return nnf, target_style
+
+
+class PyramidPatchMatcher:
+    def __init__(
+        self, image_height, image_width, channel, minimum_patch_size,
+        threads_per_block=8, num_iter=5, gpu_id=0, guide_weight=10.0,
+        use_mean_target_style=False, use_pairwise_patch_error=False,
+        tracking_window_size=0,
+        initialize="identity"
+    ):
+        maximum_patch_size = minimum_patch_size + (num_iter - 1) * 2
+        self.pyramid_level = int(np.log2(min(image_height, image_width) / maximum_patch_size))
+        self.pyramid_heights = []
+        self.pyramid_widths = []
+        self.patch_matchers = []
+        self.minimum_patch_size = minimum_patch_size
+        self.num_iter = num_iter
+        self.gpu_id = gpu_id
+        self.initialize = initialize
+        for level in range(self.pyramid_level):
+            height = image_height//(2**(self.pyramid_level - 1 - level))
+            width = image_width//(2**(self.pyramid_level - 1 - level))
+            self.pyramid_heights.append(height)
+            self.pyramid_widths.append(width)
+            self.patch_matchers.append(PatchMatcher(
+                height, width, channel, minimum_patch_size=minimum_patch_size,
+                threads_per_block=threads_per_block, num_iter=num_iter, gpu_id=gpu_id, guide_weight=guide_weight,
+                use_mean_target_style=use_mean_target_style, use_pairwise_patch_error=use_pairwise_patch_error,
+                tracking_window_size=tracking_window_size
+            ))
+
+    def resample_image(self, images, level):
+        height, width = self.pyramid_heights[level], self.pyramid_widths[level]
+        images = images.get()
+        images_resample = []
+        for image in images:
+            image_resample = cv2.resize(image, (width, height), interpolation=cv2.INTER_AREA)
+            images_resample.append(image_resample)
+        images_resample = cp.array(np.stack(images_resample), dtype=cp.float32)
+        return images_resample
+
+    def initialize_nnf(self, batch_size):
+        if self.initialize == "random":
+            height, width = self.pyramid_heights[0], self.pyramid_widths[0]
+            nnf = cp.stack([
+                cp.random.randint(0, height, (batch_size, height, width), dtype=cp.int32),
+                cp.random.randint(0, width, (batch_size, height, width), dtype=cp.int32)
+            ], axis=3)
+        elif self.initialize == "identity":
+            height, width = self.pyramid_heights[0], self.pyramid_widths[0]
+            nnf = cp.stack([
+                cp.repeat(cp.arange(height), width).reshape(height, width),
+                cp.tile(cp.arange(width), height).reshape(height, width)
+            ], axis=2)
+            nnf = cp.stack([nnf] * batch_size)
+        else:
+            raise NotImplementedError()
+        return nnf
+
+    def update_nnf(self, nnf, level):
+        # upscale
+        nnf = nnf.repeat(2, axis=1).repeat(2, axis=2) * 2
+        nnf[:,[i for i in range(nnf.shape[0]) if i&1],:,0] += 1
+        nnf[:,:,[i for i in range(nnf.shape[0]) if i&1],1] += 1
+        # check if scale is 2
+        height, width = self.pyramid_heights[level], self.pyramid_widths[level]
+        if height != nnf.shape[0] * 2 or width != nnf.shape[1] * 2:
+            nnf = nnf.get().astype(np.float32)
+            nnf = [cv2.resize(n, (width, height), interpolation=cv2.INTER_LINEAR) for n in nnf]
+            nnf = cp.array(np.stack(nnf), dtype=cp.int32)
+            nnf = self.patch_matchers[level].clamp_bound(nnf)
+        return nnf
+
+    def apply_nnf_to_image(self, nnf, image):
+        with cp.cuda.Device(self.gpu_id):
+            image = self.patch_matchers[-1].pad_image(image)
+            image = self.patch_matchers[-1].apply_nnf_to_image(nnf, image)
+        return image
+
+    def estimate_nnf(self, source_guide, target_guide, source_style):
+        with cp.cuda.Device(self.gpu_id):
+            if not isinstance(source_guide, cp.ndarray):
+                source_guide = cp.array(source_guide, dtype=cp.float32)
+            if not isinstance(target_guide, cp.ndarray):
+                target_guide = cp.array(target_guide, dtype=cp.float32)
+            if not isinstance(source_style, cp.ndarray):
+                source_style = cp.array(source_style, dtype=cp.float32)
+            for level in range(self.pyramid_level):
+                nnf = self.initialize_nnf(source_guide.shape[0]) if level==0 else self.update_nnf(nnf, level)
+                source_guide_ = self.resample_image(source_guide, level)
+                target_guide_ = self.resample_image(target_guide, level)
+                source_style_ = self.resample_image(source_style, level)
+                nnf, target_style = self.patch_matchers[level].estimate_nnf(
+                    source_guide_, target_guide_, source_style_, nnf
+                )
+        return nnf.get(), target_style.get()

diffsynth/extensions/FastBlend/runners/__init__.py

@@ -0,0 +1,4 @@
+from .accurate import AccurateModeRunner
+from .fast import FastModeRunner
+from .balanced import BalancedModeRunner
+from .interpolation import InterpolationModeRunner, InterpolationModeSingleFrameRunner

diffsynth/extensions/FastBlend/runners/accurate.py

@@ -0,0 +1,35 @@
+from ..patch_match import PyramidPatchMatcher
+import os
+import numpy as np
+from PIL import Image
+from tqdm import tqdm
+
+
+class AccurateModeRunner:
+    def __init__(self):
+        pass
+
+    def run(self, frames_guide, frames_style, batch_size, window_size, ebsynth_config, desc="Accurate Mode", save_path=None):
+        patch_match_engine = PyramidPatchMatcher(
+            image_height=frames_style[0].shape[0],
+            image_width=frames_style[0].shape[1],
+            channel=3,
+            use_mean_target_style=True,
+            **ebsynth_config
+        )
+        # run
+        n = len(frames_style)
+        for target in tqdm(range(n), desc=desc):
+            l, r = max(target - window_size, 0), min(target + window_size + 1, n)
+            remapped_frames = []
+            for i in range(l, r, batch_size):
+                j = min(i + batch_size, r)
+                source_guide = np.stack([frames_guide[source] for source in range(i, j)])
+                target_guide = np.stack([frames_guide[target]] * (j - i))
+                source_style = np.stack([frames_style[source] for source in range(i, j)])
+                _, target_style = patch_match_engine.estimate_nnf(source_guide, target_guide, source_style)
+                remapped_frames.append(target_style)
+            frame = np.concatenate(remapped_frames, axis=0).mean(axis=0)
+            frame = frame.clip(0, 255).astype("uint8")
+            if save_path is not None:
+                Image.fromarray(frame).save(os.path.join(save_path, "%05d.png" % target))

diffsynth/extensions/FastBlend/runners/balanced.py

@@ -0,0 +1,46 @@
+from ..patch_match import PyramidPatchMatcher
+import os
+import numpy as np
+from PIL import Image
+from tqdm import tqdm
+
+
+class BalancedModeRunner:
+    def __init__(self):
+        pass
+
+    def run(self, frames_guide, frames_style, batch_size, window_size, ebsynth_config, desc="Balanced Mode", save_path=None):
+        patch_match_engine = PyramidPatchMatcher(
+            image_height=frames_style[0].shape[0],
+            image_width=frames_style[0].shape[1],
+            channel=3,
+            **ebsynth_config
+        )
+        # tasks
+        n = len(frames_style)
+        tasks = []
+        for target in range(n):
+            for source in range(target - window_size, target + window_size + 1):
+                if source >= 0 and source < n and source != target:
+                    tasks.append((source, target))
+        # run
+        frames = [(None, 1) for i in range(n)]
+        for batch_id in tqdm(range(0, len(tasks), batch_size), desc=desc):
+            tasks_batch = tasks[batch_id: min(batch_id+batch_size, len(tasks))]
+            source_guide = np.stack([frames_guide[source] for source, target in tasks_batch])
+            target_guide = np.stack([frames_guide[target] for source, target in tasks_batch])
+            source_style = np.stack([frames_style[source] for source, target in tasks_batch])
+            _, target_style = patch_match_engine.estimate_nnf(source_guide, target_guide, source_style)
+            for (source, target), result in zip(tasks_batch, target_style):
+                frame, weight = frames[target]
+                if frame is None:
+                    frame = frames_style[target]
+                frames[target] = (
+                    frame * (weight / (weight + 1)) + result / (weight + 1),
+                    weight + 1
+                )
+                if weight + 1 == min(n, target + window_size + 1) - max(0, target - window_size):
+                    frame = frame.clip(0, 255).astype("uint8")
+                    if save_path is not None:
+                        Image.fromarray(frame).save(os.path.join(save_path, "%05d.png" % target))
+                    frames[target] = (None, 1)

diffsynth/extensions/FastBlend/runners/fast.py

@@ -0,0 +1,141 @@
+from ..patch_match import PyramidPatchMatcher
+import functools, os
+import numpy as np
+from PIL import Image
+from tqdm import tqdm
+
+
+class TableManager:
+    def __init__(self):
+        pass
+
+    def task_list(self, n):
+        tasks = []
+        max_level = 1
+        while (1<<max_level)<=n:
+            max_level += 1
+        for i in range(n):
+            j = i
+            for level in range(max_level):
+                if i&(1<<level):
+                    continue
+                j |= 1<<level
+                if j>=n:
+                    break
+                meta_data = {
+                    "source": i,
+                    "target": j,
+                    "level": level + 1
+                }
+                tasks.append(meta_data)
+        tasks.sort(key=functools.cmp_to_key(lambda u, v: u["level"]-v["level"]))
+        return tasks
+    
+    def build_remapping_table(self, frames_guide, frames_style, patch_match_engine, batch_size, desc=""):
+        n = len(frames_guide)
+        tasks = self.task_list(n)
+        remapping_table = [[(frames_style[i], 1)] for i in range(n)]
+        for batch_id in tqdm(range(0, len(tasks), batch_size), desc=desc):
+            tasks_batch = tasks[batch_id: min(batch_id+batch_size, len(tasks))]
+            source_guide = np.stack([frames_guide[task["source"]] for task in tasks_batch])
+            target_guide = np.stack([frames_guide[task["target"]] for task in tasks_batch])
+            source_style = np.stack([frames_style[task["source"]] for task in tasks_batch])
+            _, target_style = patch_match_engine.estimate_nnf(source_guide, target_guide, source_style)
+            for task, result in zip(tasks_batch, target_style):
+                target, level = task["target"], task["level"]
+                if len(remapping_table[target])==level:
+                    remapping_table[target].append((result, 1))
+                else:
+                    frame, weight = remapping_table[target][level]
+                    remapping_table[target][level] = (
+                        frame * (weight / (weight + 1)) + result / (weight + 1),
+                        weight + 1
+                    )
+        return remapping_table
+
+    def remapping_table_to_blending_table(self, table):
+        for i in range(len(table)):
+            for j in range(1, len(table[i])):
+                frame_1, weight_1 = table[i][j-1]
+                frame_2, weight_2 = table[i][j]
+                frame = (frame_1 + frame_2) / 2
+                weight = weight_1 + weight_2
+                table[i][j] = (frame, weight)
+        return table
+
+    def tree_query(self, leftbound, rightbound):
+        node_list = []
+        node_index = rightbound
+        while node_index>=leftbound:
+            node_level = 0
+            while (1<<node_level)&node_index and node_index-(1<<node_level+1)+1>=leftbound:
+                node_level += 1
+            node_list.append((node_index, node_level))
+            node_index -= 1<<node_level
+        return node_list
+
+    def process_window_sum(self, frames_guide, blending_table, patch_match_engine, window_size, batch_size, desc=""):
+        n = len(blending_table)
+        tasks = []
+        frames_result = []
+        for target in range(n):
+            node_list = self.tree_query(max(target-window_size, 0), target)
+            for source, level in node_list:
+                if source!=target:
+                    meta_data = {
+                        "source": source,
+                        "target": target,
+                        "level": level
+                    }
+                    tasks.append(meta_data)
+                else:
+                    frames_result.append(blending_table[target][level])
+        for batch_id in tqdm(range(0, len(tasks), batch_size), desc=desc):
+            tasks_batch = tasks[batch_id: min(batch_id+batch_size, len(tasks))]
+            source_guide = np.stack([frames_guide[task["source"]] for task in tasks_batch])
+            target_guide = np.stack([frames_guide[task["target"]] for task in tasks_batch])
+            source_style = np.stack([blending_table[task["source"]][task["level"]][0] for task in tasks_batch])
+            _, target_style = patch_match_engine.estimate_nnf(source_guide, target_guide, source_style)
+            for task, frame_2 in zip(tasks_batch, target_style):
+                source, target, level = task["source"], task["target"], task["level"]
+                frame_1, weight_1 = frames_result[target]
+                weight_2 = blending_table[source][level][1]
+                weight = weight_1 + weight_2
+                frame = frame_1 * (weight_1 / weight) + frame_2 * (weight_2 / weight)
+                frames_result[target] = (frame, weight)
+        return frames_result
+
+
+class FastModeRunner:
+    def __init__(self):
+        pass
+
+    def run(self, frames_guide, frames_style, batch_size, window_size, ebsynth_config, save_path=None):
+        frames_guide = frames_guide.raw_data()
+        frames_style = frames_style.raw_data()
+        table_manager = TableManager()
+        patch_match_engine = PyramidPatchMatcher(
+            image_height=frames_style[0].shape[0],
+            image_width=frames_style[0].shape[1],
+            channel=3,
+            **ebsynth_config
+        )
+        # left part
+        table_l = table_manager.build_remapping_table(frames_guide, frames_style, patch_match_engine, batch_size, desc="Fast Mode Step 1/4")
+        table_l = table_manager.remapping_table_to_blending_table(table_l)
+        table_l = table_manager.process_window_sum(frames_guide, table_l, patch_match_engine, window_size, batch_size, desc="Fast Mode Step 2/4")
+        # right part
+        table_r = table_manager.build_remapping_table(frames_guide[::-1], frames_style[::-1], patch_match_engine, batch_size, desc="Fast Mode Step 3/4")
+        table_r = table_manager.remapping_table_to_blending_table(table_r)
+        table_r = table_manager.process_window_sum(frames_guide[::-1], table_r, patch_match_engine, window_size, batch_size, desc="Fast Mode Step 4/4")[::-1]
+        # merge
+        frames = []
+        for (frame_l, weight_l), frame_m, (frame_r, weight_r) in zip(table_l, frames_style, table_r):
+            weight_m = -1
+            weight = weight_l + weight_m + weight_r
+            frame = frame_l * (weight_l / weight) + frame_m * (weight_m / weight) + frame_r * (weight_r / weight)
+            frames.append(frame)
+        frames = [frame.clip(0, 255).astype("uint8") for frame in frames]
+        if save_path is not None:
+            for target, frame in enumerate(frames):
+                Image.fromarray(frame).save(os.path.join(save_path, "%05d.png" % target))

diffsynth/extensions/FastBlend/runners/interpolation.py

@@ -0,0 +1,121 @@
+from ..patch_match import PyramidPatchMatcher
+import os
+import numpy as np
+from PIL import Image
+from tqdm import tqdm
+
+
+class InterpolationModeRunner:
+    def __init__(self):
+        pass
+
+    def get_index_dict(self, index_style):
+        index_dict = {}
+        for i, index in enumerate(index_style):
+            index_dict[index] = i
+        return index_dict
+
+    def get_weight(self, l, m, r):
+        weight_l, weight_r = abs(m - r), abs(m - l)
+        if weight_l + weight_r == 0:
+            weight_l, weight_r = 0.5, 0.5
+        else:
+            weight_l, weight_r = weight_l / (weight_l + weight_r), weight_r / (weight_l + weight_r)
+        return weight_l, weight_r
+
+    def get_task_group(self, index_style, n):
+        task_group = []
+        index_style = sorted(index_style)
+        # first frame
+        if index_style[0]>0:
+            tasks = []
+            for m in range(index_style[0]):
+                tasks.append((index_style[0], m, index_style[0]))
+            task_group.append(tasks)
+        # middle frames
+        for l, r in zip(index_style[:-1], index_style[1:]):
+            tasks = []
+            for m in range(l, r):
+                tasks.append((l, m, r))
+            task_group.append(tasks)
+        # last frame
+        tasks = []
+        for m in range(index_style[-1], n):
+            tasks.append((index_style[-1], m, index_style[-1]))
+        task_group.append(tasks)
+        return task_group
+
+    def run(self, frames_guide, frames_style, index_style, batch_size, ebsynth_config, save_path=None):
+        patch_match_engine = PyramidPatchMatcher(
+            image_height=frames_style[0].shape[0],
+            image_width=frames_style[0].shape[1],
+            channel=3,
+            use_mean_target_style=False,
+            use_pairwise_patch_error=True,
+            **ebsynth_config
+        )
+        # task
+        index_dict = self.get_index_dict(index_style)
+        task_group = self.get_task_group(index_style, len(frames_guide))
+        # run
+        for tasks in task_group:
+            index_start, index_end = min([i[1] for i in tasks]), max([i[1] for i in tasks])
+            for batch_id in tqdm(range(0, len(tasks), batch_size), desc=f"Rendering frames {index_start}...{index_end}"):
+                tasks_batch = tasks[batch_id: min(batch_id+batch_size, len(tasks))]
+                source_guide, target_guide, source_style = [], [], []
+                for l, m, r in tasks_batch:
+                    # l -> m
+                    source_guide.append(frames_guide[l])
+                    target_guide.append(frames_guide[m])
+                    source_style.append(frames_style[index_dict[l]])
+                    # r -> m
+                    source_guide.append(frames_guide[r])
+                    target_guide.append(frames_guide[m])
+                    source_style.append(frames_style[index_dict[r]])
+                source_guide = np.stack(source_guide)
+                target_guide = np.stack(target_guide)
+                source_style = np.stack(source_style)
+                _, target_style = patch_match_engine.estimate_nnf(source_guide, target_guide, source_style)
+                if save_path is not None:
+                    for frame_l, frame_r, (l, m, r) in zip(target_style[0::2], target_style[1::2], tasks_batch):
+                        weight_l, weight_r = self.get_weight(l, m, r)
+                        frame = frame_l * weight_l + frame_r * weight_r
+                        frame = frame.clip(0, 255).astype("uint8")
+                        Image.fromarray(frame).save(os.path.join(save_path, "%05d.png" % m))
+
+
+class InterpolationModeSingleFrameRunner:
+    def __init__(self):
+        pass
+
+    def run(self, frames_guide, frames_style, index_style, batch_size, ebsynth_config, save_path=None):
+        # check input
+        tracking_window_size = ebsynth_config["tracking_window_size"]
+        if tracking_window_size * 2 >= batch_size:
+            raise ValueError("batch_size should be larger than track_window_size * 2")
+        frame_style = frames_style[0]
+        frame_guide = frames_guide[index_style[0]]
+        patch_match_engine = PyramidPatchMatcher(
+            image_height=frame_style.shape[0],
+            image_width=frame_style.shape[1],
+            channel=3,
+            **ebsynth_config
+        )
+        # run
+        frame_id, n = 0, len(frames_guide)
+        for i in tqdm(range(0, n, batch_size - tracking_window_size * 2), desc=f"Rendering frames 0...{n}"):
+            if i + batch_size > n:
+                l, r = max(n - batch_size, 0), n
+            else:
+                l, r = i, i + batch_size
+            source_guide = np.stack([frame_guide] * (r-l))
+            target_guide = np.stack([frames_guide[i] for i in range(l, r)])
+            source_style = np.stack([frame_style] * (r-l))
+            _, target_style = patch_match_engine.estimate_nnf(source_guide, target_guide, source_style)
+            for i, frame in zip(range(l, r), target_style):
+                if i==frame_id:
+                    frame = frame.clip(0, 255).astype("uint8")
+                    Image.fromarray(frame).save(os.path.join(save_path, "%05d.png" % frame_id))
+                    frame_id += 1
+                if r < n and r-frame_id <= tracking_window_size:
+                    break

diffsynth/extensions/ImageQualityMetric/BLIP/__init__.py

@@ -0,0 +1 @@
+from .blip_pretrain import *

diffsynth/extensions/ImageQualityMetric/BLIP/blip.py

@@ -0,0 +1,77 @@
+'''
+ * Adapted from BLIP (https://github.com/salesforce/BLIP)
+'''
+
+import warnings
+warnings.filterwarnings("ignore")
+
+import torch
+import os
+from urllib.parse import urlparse
+from timm.models.hub import download_cached_file
+from transformers import BertTokenizer
+from .vit import VisionTransformer, interpolate_pos_embed
+
+
+def default_bert():
+    current_dir = os.path.dirname(os.path.abspath(__file__))
+    project_root = os.path.abspath(os.path.join(current_dir, '../../../../'))
+    model_path = os.path.join(project_root, 'models', 'QualityMetric')
+    return os.path.join(model_path, "bert-base-uncased")
+
+
+def init_tokenizer(bert_model_path):
+    tokenizer = BertTokenizer.from_pretrained(bert_model_path)
+    tokenizer.add_special_tokens({'bos_token':'[DEC]'})
+    tokenizer.add_special_tokens({'additional_special_tokens':['[ENC]']})       
+    tokenizer.enc_token_id = tokenizer.additional_special_tokens_ids[0]  
+    return tokenizer
+
+
+def create_vit(vit, image_size, use_grad_checkpointing=False, ckpt_layer=0, drop_path_rate=0):
+        
+    assert vit in ['base', 'large'], "vit parameter must be base or large"
+    if vit=='base':
+        vision_width = 768
+        visual_encoder = VisionTransformer(img_size=image_size, patch_size=16, embed_dim=vision_width, depth=12, 
+                                           num_heads=12, use_grad_checkpointing=use_grad_checkpointing, ckpt_layer=ckpt_layer,
+                                           drop_path_rate=0 or drop_path_rate
+                                          )   
+    elif vit=='large':
+        vision_width = 1024
+        visual_encoder = VisionTransformer(img_size=image_size, patch_size=16, embed_dim=vision_width, depth=24, 
+                                           num_heads=16, use_grad_checkpointing=use_grad_checkpointing, ckpt_layer=ckpt_layer,
+                                           drop_path_rate=0.1 or drop_path_rate
+                                          )   
+    return visual_encoder, vision_width
+
+
+def is_url(url_or_filename):
+    parsed = urlparse(url_or_filename)
+    return parsed.scheme in ("http", "https")
+
+def load_checkpoint(model,url_or_filename):
+    if is_url(url_or_filename):
+        cached_file = download_cached_file(url_or_filename, check_hash=False, progress=True)
+        checkpoint = torch.load(cached_file, map_location='cpu') 
+    elif os.path.isfile(url_or_filename):        
+        checkpoint = torch.load(url_or_filename, map_location='cpu') 
+    else:
+        raise RuntimeError('checkpoint url or path is invalid')
+        
+    state_dict = checkpoint['model']
+    
+    state_dict['visual_encoder.pos_embed'] = interpolate_pos_embed(state_dict['visual_encoder.pos_embed'],model.visual_encoder) 
+    if 'visual_encoder_m.pos_embed' in model.state_dict().keys():
+        state_dict['visual_encoder_m.pos_embed'] = interpolate_pos_embed(state_dict['visual_encoder_m.pos_embed'],
+                                                                         model.visual_encoder_m)    
+    for key in model.state_dict().keys():
+        if key in state_dict.keys():
+            if state_dict[key].shape!=model.state_dict()[key].shape:
+                print(key, ": ", state_dict[key].shape, ', ', model.state_dict()[key].shape)
+                del state_dict[key]
+    
+    msg = model.load_state_dict(state_dict,strict=False)
+    print('load checkpoint from %s'%url_or_filename)  
+    return model,msg
+    

diffsynth/extensions/ImageQualityMetric/BLIP/blip_pretrain.py

@@ -0,0 +1,44 @@
+'''
+ * Adapted from BLIP (https://github.com/salesforce/BLIP)
+'''
+
+import transformers
+transformers.logging.set_verbosity_error()
+
+from torch import nn
+import os
+from .med import BertConfig, BertModel
+from .blip import create_vit, init_tokenizer
+
+class BLIP_Pretrain(nn.Module):
+    def __init__(self,                 
+                 med_config = "med_config.json",  
+                 image_size = 224,
+                 vit = 'base',
+                 vit_grad_ckpt = False,
+                 vit_ckpt_layer = 0,                    
+                 embed_dim = 256,     
+                 queue_size = 57600,
+                 momentum = 0.995,
+                 bert_model_path = ""
+                 ):
+        """
+        Args:
+            med_config (str): path for the mixture of encoder-decoder model's configuration file
+            image_size (int): input image size
+            vit (str): model size of vision transformer
+        """               
+        super().__init__()
+        
+        self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer, 0)
+        
+        self.tokenizer = init_tokenizer(bert_model_path)   
+        encoder_config = BertConfig.from_json_file(med_config)
+        encoder_config.encoder_width = vision_width
+        self.text_encoder = BertModel(config=encoder_config, add_pooling_layer=False)
+
+        text_width = self.text_encoder.config.hidden_size
+        
+        self.vision_proj = nn.Linear(vision_width, embed_dim)
+        self.text_proj = nn.Linear(text_width, embed_dim)
+

diffsynth/extensions/ImageQualityMetric/BLIP/med.py

@@ -0,0 +1,947 @@
+'''
+ * Adapted from BLIP (https://github.com/salesforce/BLIP)
+ * Based on huggingface code base
+ * https://github.com/huggingface/transformers/blob/v4.15.0/src/transformers/models/bert
+'''
+
+import math
+from typing import Tuple
+
+import torch
+from torch import Tensor, device, nn
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+
+from transformers.activations import ACT2FN
+from transformers.file_utils import (
+    ModelOutput,
+)
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPastAndCrossAttentions,
+    BaseModelOutputWithPoolingAndCrossAttentions,
+    CausalLMOutputWithCrossAttentions,
+    MaskedLMOutput,
+    MultipleChoiceModelOutput,
+    NextSentencePredictorOutput,
+    QuestionAnsweringModelOutput,
+    SequenceClassifierOutput,
+    TokenClassifierOutput,
+)
+from transformers.modeling_utils import (
+    PreTrainedModel,
+    apply_chunking_to_forward,
+    find_pruneable_heads_and_indices,
+    prune_linear_layer,
+)
+from transformers.utils import logging
+from transformers.models.bert.configuration_bert import BertConfig
+
+
+logger = logging.get_logger(__name__)
+
+
+class BertEmbeddings(nn.Module):
+    """Construct the embeddings from word and position embeddings."""
+
+    def __init__(self, config):
+        super().__init__()
+        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
+        self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)
+
+        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
+        # any TensorFlow checkpoint file
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+        # position_ids (1, len position emb) is contiguous in memory and exported when serialized
+        self.register_buffer("position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)))
+        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
+        
+        self.config = config
+
+    def forward(
+        self, input_ids=None, position_ids=None, inputs_embeds=None, past_key_values_length=0
+    ):
+        if input_ids is not None:
+            input_shape = input_ids.size()
+        else:
+            input_shape = inputs_embeds.size()[:-1]
+
+        seq_length = input_shape[1]
+
+        if position_ids is None:
+            position_ids = self.position_ids[:, past_key_values_length : seq_length + past_key_values_length]
+
+        if inputs_embeds is None:
+            inputs_embeds = self.word_embeddings(input_ids)
+
+        embeddings = inputs_embeds
+
+        if self.position_embedding_type == "absolute":
+            position_embeddings = self.position_embeddings(position_ids)
+            embeddings += position_embeddings
+        embeddings = self.LayerNorm(embeddings)
+        embeddings = self.dropout(embeddings)
+        return embeddings
+
+
+class BertSelfAttention(nn.Module):
+    def __init__(self, config, is_cross_attention):
+        super().__init__()
+        self.config = config
+        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
+            raise ValueError(
+                "The hidden size (%d) is not a multiple of the number of attention "
+                "heads (%d)" % (config.hidden_size, config.num_attention_heads)
+            )
+        
+        self.num_attention_heads = config.num_attention_heads
+        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
+        self.all_head_size = self.num_attention_heads * self.attention_head_size
+
+        self.query = nn.Linear(config.hidden_size, self.all_head_size)
+        if is_cross_attention:
+            self.key = nn.Linear(config.encoder_width, self.all_head_size)
+            self.value = nn.Linear(config.encoder_width, self.all_head_size)
+        else:
+            self.key = nn.Linear(config.hidden_size, self.all_head_size)
+            self.value = nn.Linear(config.hidden_size, self.all_head_size)
+
+        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
+        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
+        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
+            self.max_position_embeddings = config.max_position_embeddings
+            self.distance_embedding = nn.Embedding(2 * config.max_position_embeddings - 1, self.attention_head_size)
+        self.save_attention = False   
+            
+    def save_attn_gradients(self, attn_gradients):
+        self.attn_gradients = attn_gradients
+        
+    def get_attn_gradients(self):
+        return self.attn_gradients
+    
+    def save_attention_map(self, attention_map):
+        self.attention_map = attention_map
+        
+    def get_attention_map(self):
+        return self.attention_map
+    
+    def transpose_for_scores(self, x):
+        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
+        x = x.view(*new_x_shape)
+        return x.permute(0, 2, 1, 3)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+    ):
+        mixed_query_layer = self.query(hidden_states)
+
+        # If this is instantiated as a cross-attention module, the keys
+        # and values come from an encoder; the attention mask needs to be
+        # such that the encoder's padding tokens are not attended to.
+        is_cross_attention = encoder_hidden_states is not None
+
+        if is_cross_attention:
+            key_layer = self.transpose_for_scores(self.key(encoder_hidden_states))
+            value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))
+            attention_mask = encoder_attention_mask
+        elif past_key_value is not None:
+            key_layer = self.transpose_for_scores(self.key(hidden_states))
+            value_layer = self.transpose_for_scores(self.value(hidden_states))
+            key_layer = torch.cat([past_key_value[0], key_layer], dim=2)
+            value_layer = torch.cat([past_key_value[1], value_layer], dim=2)
+        else:
+            key_layer = self.transpose_for_scores(self.key(hidden_states))
+            value_layer = self.transpose_for_scores(self.value(hidden_states))
+
+        query_layer = self.transpose_for_scores(mixed_query_layer)
+
+        past_key_value = (key_layer, value_layer)
+
+        # Take the dot product between "query" and "key" to get the raw attention scores.
+        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
+
+        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
+            seq_length = hidden_states.size()[1]
+            position_ids_l = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(-1, 1)
+            position_ids_r = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(1, -1)
+            distance = position_ids_l - position_ids_r
+            positional_embedding = self.distance_embedding(distance + self.max_position_embeddings - 1)
+            positional_embedding = positional_embedding.to(dtype=query_layer.dtype)  # fp16 compatibility
+
+            if self.position_embedding_type == "relative_key":
+                relative_position_scores = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
+                attention_scores = attention_scores + relative_position_scores
+            elif self.position_embedding_type == "relative_key_query":
+                relative_position_scores_query = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
+                relative_position_scores_key = torch.einsum("bhrd,lrd->bhlr", key_layer, positional_embedding)
+                attention_scores = attention_scores + relative_position_scores_query + relative_position_scores_key
+
+        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
+        if attention_mask is not None:
+            # Apply the attention mask is (precomputed for all layers in BertModel forward() function)
+            attention_scores = attention_scores + attention_mask
+
+        # Normalize the attention scores to probabilities.
+        attention_probs = nn.Softmax(dim=-1)(attention_scores)
+        
+        if is_cross_attention and self.save_attention:
+            self.save_attention_map(attention_probs)
+            attention_probs.register_hook(self.save_attn_gradients)         
+
+        # This is actually dropping out entire tokens to attend to, which might
+        # seem a bit unusual, but is taken from the original Transformer paper.
+        attention_probs_dropped = self.dropout(attention_probs)
+
+        # Mask heads if we want to
+        if head_mask is not None:
+            attention_probs_dropped = attention_probs_dropped * head_mask
+
+        context_layer = torch.matmul(attention_probs_dropped, value_layer)
+
+        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
+        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
+        context_layer = context_layer.view(*new_context_layer_shape)
+
+        outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)
+
+        outputs = outputs + (past_key_value,)
+        return outputs
+
+
+class BertSelfOutput(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, hidden_states, input_tensor):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+class BertAttention(nn.Module):
+    def __init__(self, config, is_cross_attention=False):
+        super().__init__()
+        self.self = BertSelfAttention(config, is_cross_attention)
+        self.output = BertSelfOutput(config)
+        self.pruned_heads = set()
+
+    def prune_heads(self, heads):
+        if len(heads) == 0:
+            return
+        heads, index = find_pruneable_heads_and_indices(
+            heads, self.self.num_attention_heads, self.self.attention_head_size, self.pruned_heads
+        )
+
+        # Prune linear layers
+        self.self.query = prune_linear_layer(self.self.query, index)
+        self.self.key = prune_linear_layer(self.self.key, index)
+        self.self.value = prune_linear_layer(self.self.value, index)
+        self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)
+
+        # Update hyper params and store pruned heads
+        self.self.num_attention_heads = self.self.num_attention_heads - len(heads)
+        self.self.all_head_size = self.self.attention_head_size * self.self.num_attention_heads
+        self.pruned_heads = self.pruned_heads.union(heads)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+    ):
+        self_outputs = self.self(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            encoder_hidden_states,
+            encoder_attention_mask,
+            past_key_value,
+            output_attentions,
+        )
+        attention_output = self.output(self_outputs[0], hidden_states)
+        outputs = (attention_output,) + self_outputs[1:]  # add attentions if we output them
+        return outputs
+
+
+class BertIntermediate(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
+        if isinstance(config.hidden_act, str):
+            self.intermediate_act_fn = ACT2FN[config.hidden_act]
+        else:
+            self.intermediate_act_fn = config.hidden_act
+
+    def forward(self, hidden_states):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.intermediate_act_fn(hidden_states)
+        return hidden_states
+
+
+class BertOutput(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, hidden_states, input_tensor):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+class BertLayer(nn.Module):
+    def __init__(self, config, layer_num):
+        super().__init__()
+        self.config = config
+        self.chunk_size_feed_forward = config.chunk_size_feed_forward
+        self.seq_len_dim = 1
+        self.attention = BertAttention(config)      
+        self.layer_num = layer_num          
+        if self.config.add_cross_attention:
+            self.crossattention = BertAttention(config, is_cross_attention=self.config.add_cross_attention)
+        self.intermediate = BertIntermediate(config)
+        self.output = BertOutput(config)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+        mode=None,
+    ):
+        # decoder uni-directional self-attention cached key/values tuple is at positions 1,2
+        self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None
+        self_attention_outputs = self.attention(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            output_attentions=output_attentions,
+            past_key_value=self_attn_past_key_value,
+        )
+        attention_output = self_attention_outputs[0]
+
+        outputs = self_attention_outputs[1:-1]
+        present_key_value = self_attention_outputs[-1]
+
+        if mode=='multimodal':
+            assert encoder_hidden_states is not None, "encoder_hidden_states must be given for cross-attention layers"
+
+            cross_attention_outputs = self.crossattention(
+                attention_output,
+                attention_mask,
+                head_mask,
+                encoder_hidden_states,
+                encoder_attention_mask,
+                output_attentions=output_attentions,
+            )
+            attention_output = cross_attention_outputs[0]
+            outputs = outputs + cross_attention_outputs[1:-1]  # add cross attentions if we output attention weights                               
+        layer_output = apply_chunking_to_forward(
+            self.feed_forward_chunk, self.chunk_size_feed_forward, self.seq_len_dim, attention_output
+        )
+        outputs = (layer_output,) + outputs
+
+        outputs = outputs + (present_key_value,)
+
+        return outputs
+
+    def feed_forward_chunk(self, attention_output):
+        intermediate_output = self.intermediate(attention_output)
+        layer_output = self.output(intermediate_output, attention_output)
+        return layer_output
+
+
+class BertEncoder(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.layer = nn.ModuleList([BertLayer(config,i) for i in range(config.num_hidden_layers)])
+        self.gradient_checkpointing = False
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=False,
+        output_hidden_states=False,
+        return_dict=True,
+        mode='multimodal',
+    ):
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attentions = () if output_attentions else None
+        all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
+
+        next_decoder_cache = () if use_cache else None
+               
+        for i in range(self.config.num_hidden_layers):
+            layer_module = self.layer[i]
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            layer_head_mask = head_mask[i] if head_mask is not None else None
+            past_key_value = past_key_values[i] if past_key_values is not None else None
+
+            if self.gradient_checkpointing and self.training:
+
+                if use_cache:
+                    logger.warn(
+                        "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                    )
+                    use_cache = False
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        return module(*inputs, past_key_value, output_attentions)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(layer_module),
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    mode=mode,
+                )
+            else:
+                layer_outputs = layer_module(
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    past_key_value,
+                    output_attentions,
+                    mode=mode,
+                )
+
+            hidden_states = layer_outputs[0]
+            if use_cache:
+                next_decoder_cache += (layer_outputs[-1],)
+            if output_attentions:
+                all_self_attentions = all_self_attentions + (layer_outputs[1],)
+
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(
+                v
+                for v in [
+                    hidden_states,
+                    next_decoder_cache,
+                    all_hidden_states,
+                    all_self_attentions,
+                    all_cross_attentions,
+                ]
+                if v is not None
+            )
+        return BaseModelOutputWithPastAndCrossAttentions(
+            last_hidden_state=hidden_states,
+            past_key_values=next_decoder_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+            cross_attentions=all_cross_attentions,
+        )
+
+
+class BertPooler(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.activation = nn.Tanh()
+
+    def forward(self, hidden_states):
+        # We "pool" the model by simply taking the hidden state corresponding
+        # to the first token.
+        first_token_tensor = hidden_states[:, 0]
+        pooled_output = self.dense(first_token_tensor)
+        pooled_output = self.activation(pooled_output)
+        return pooled_output
+
+
+class BertPredictionHeadTransform(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        if isinstance(config.hidden_act, str):
+            self.transform_act_fn = ACT2FN[config.hidden_act]
+        else:
+            self.transform_act_fn = config.hidden_act
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+
+    def forward(self, hidden_states):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.transform_act_fn(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states)
+        return hidden_states
+
+
+class BertLMPredictionHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.transform = BertPredictionHeadTransform(config)
+
+        # The output weights are the same as the input embeddings, but there is
+        # an output-only bias for each token.
+        self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        self.bias = nn.Parameter(torch.zeros(config.vocab_size))
+
+        # Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`
+        self.decoder.bias = self.bias
+
+    def forward(self, hidden_states):
+        hidden_states = self.transform(hidden_states)
+        hidden_states = self.decoder(hidden_states)
+        return hidden_states
+
+
+class BertOnlyMLMHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.predictions = BertLMPredictionHead(config)
+
+    def forward(self, sequence_output):
+        prediction_scores = self.predictions(sequence_output)
+        return prediction_scores
+
+
+class BertPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = BertConfig
+    base_model_prefix = "bert"
+    _keys_to_ignore_on_load_missing = [r"position_ids"]
+
+    def _init_weights(self, module):
+        """ Initialize the weights """
+        if isinstance(module, (nn.Linear, nn.Embedding)):
+            # Slightly different from the TF version which uses truncated_normal for initialization
+            # cf https://github.com/pytorch/pytorch/pull/5617
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+        elif isinstance(module, nn.LayerNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+        if isinstance(module, nn.Linear) and module.bias is not None:
+            module.bias.data.zero_()
+
+
+class BertModel(BertPreTrainedModel):
+    """
+    The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of
+    cross-attention is added between the self-attention layers, following the architecture described in `Attention is
+    all you need <https://arxiv.org/abs/1706.03762>`__ by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,
+    Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.
+    argument and :obj:`add_cross_attention` set to :obj:`True`; an :obj:`encoder_hidden_states` is then expected as an
+    input to the forward pass.
+    """
+
+    def __init__(self, config, add_pooling_layer=True):
+        super().__init__(config)
+        self.config = config
+
+        self.embeddings = BertEmbeddings(config)
+        
+        self.encoder = BertEncoder(config)
+
+        self.pooler = BertPooler(config) if add_pooling_layer else None
+
+        self.init_weights()
+ 
+
+    def get_input_embeddings(self):
+        return self.embeddings.word_embeddings
+
+    def set_input_embeddings(self, value):
+        self.embeddings.word_embeddings = value
+
+    def _prune_heads(self, heads_to_prune):
+        """
+        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
+        class PreTrainedModel
+        """
+        for layer, heads in heads_to_prune.items():
+            self.encoder.layer[layer].attention.prune_heads(heads)
+
+    
+    def get_extended_attention_mask(self, attention_mask: Tensor, input_shape: Tuple[int], device: device, is_decoder: bool) -> Tensor:
+        """
+        Makes broadcastable attention and causal masks so that future and masked tokens are ignored.
+
+        Arguments:
+            attention_mask (:obj:`torch.Tensor`):
+                Mask with ones indicating tokens to attend to, zeros for tokens to ignore.
+            input_shape (:obj:`Tuple[int]`):
+                The shape of the input to the model.
+            device: (:obj:`torch.device`):
+                The device of the input to the model.
+
+        Returns:
+            :obj:`torch.Tensor` The extended attention mask, with a the same dtype as :obj:`attention_mask.dtype`.
+        """
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        if attention_mask.dim() == 3:
+            extended_attention_mask = attention_mask[:, None, :, :]
+        elif attention_mask.dim() == 2:
+            # Provided a padding mask of dimensions [batch_size, seq_length]
+            # - if the model is a decoder, apply a causal mask in addition to the padding mask
+            # - if the model is an encoder, make the mask broadcastable to [batch_size, num_heads, seq_length, seq_length]
+            if is_decoder:
+                batch_size, seq_length = input_shape
+
+                seq_ids = torch.arange(seq_length, device=device)
+                causal_mask = seq_ids[None, None, :].repeat(batch_size, seq_length, 1) <= seq_ids[None, :, None]
+                # in case past_key_values are used we need to add a prefix ones mask to the causal mask
+                # causal and attention masks must have same type with pytorch version < 1.3
+                causal_mask = causal_mask.to(attention_mask.dtype)
+   
+                if causal_mask.shape[1] < attention_mask.shape[1]:
+                    prefix_seq_len = attention_mask.shape[1] - causal_mask.shape[1]
+                    causal_mask = torch.cat(
+                        [
+                            torch.ones((batch_size, seq_length, prefix_seq_len), device=device, dtype=causal_mask.dtype),
+                            causal_mask,
+                        ],
+                        axis=-1,
+                    )                     
+
+                extended_attention_mask = causal_mask[:, None, :, :] * attention_mask[:, None, None, :]
+            else:
+                extended_attention_mask = attention_mask[:, None, None, :]
+        else:
+            raise ValueError(
+                "Wrong shape for input_ids (shape {}) or attention_mask (shape {})".format(
+                    input_shape, attention_mask.shape
+                )
+            )
+
+        # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
+        # masked positions, this operation will create a tensor which is 0.0 for
+        # positions we want to attend and -10000.0 for masked positions.
+        # Since we are adding it to the raw scores before the softmax, this is
+        # effectively the same as removing these entirely.
+        extended_attention_mask = extended_attention_mask.to(dtype=self.dtype)  # fp16 compatibility
+        extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
+        return extended_attention_mask
+    
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        encoder_embeds=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+        is_decoder=False,
+        mode='multimodal',
+    ):
+        r"""
+        encoder_hidden_states  (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
+            the model is configured as a decoder.
+        encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
+            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
+            the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
+            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
+            If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
+            (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
+            instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
+        use_cache (:obj:`bool`, `optional`):
+            If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
+            decoding (see :obj:`past_key_values`).
+        """
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if is_decoder:
+            use_cache = use_cache if use_cache is not None else self.config.use_cache
+        else:
+            use_cache = False
+
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        elif input_ids is not None:
+            input_shape = input_ids.size()
+            batch_size, seq_length = input_shape
+            device = input_ids.device
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.size()[:-1]
+            batch_size, seq_length = input_shape
+            device = inputs_embeds.device
+        elif encoder_embeds is not None:    
+            input_shape = encoder_embeds.size()[:-1]
+            batch_size, seq_length = input_shape 
+            device = encoder_embeds.device
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds or encoder_embeds")
+
+        # past_key_values_length
+        past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0
+
+        if attention_mask is None:
+            attention_mask = torch.ones(((batch_size, seq_length + past_key_values_length)), device=device)
+            
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, 
+                                                                                 device, is_decoder)
+
+        # If a 2D or 3D attention mask is provided for the cross-attention
+        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
+        if encoder_hidden_states is not None:
+            if type(encoder_hidden_states) == list:
+                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states[0].size()
+            else:
+                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
+            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
+            
+            if type(encoder_attention_mask) == list:
+                encoder_extended_attention_mask = [self.invert_attention_mask(mask) for mask in encoder_attention_mask]
+            elif encoder_attention_mask is None:
+                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
+                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+            else:    
+                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+        else:
+            encoder_extended_attention_mask = None
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x n_heads x N x N
+        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
+        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
+        head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
+        
+        if encoder_embeds is None:
+            embedding_output = self.embeddings(
+                input_ids=input_ids,
+                position_ids=position_ids,
+                inputs_embeds=inputs_embeds,
+                past_key_values_length=past_key_values_length,
+            )
+        else:
+            embedding_output = encoder_embeds
+            
+        encoder_outputs = self.encoder(
+            embedding_output,
+            attention_mask=extended_attention_mask,
+            head_mask=head_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_extended_attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            mode=mode,
+        )
+        sequence_output = encoder_outputs[0]
+        pooled_output = self.pooler(sequence_output) if self.pooler is not None else None
+
+        if not return_dict:
+            return (sequence_output, pooled_output) + encoder_outputs[1:]
+
+        return BaseModelOutputWithPoolingAndCrossAttentions(
+            last_hidden_state=sequence_output,
+            pooler_output=pooled_output,
+            past_key_values=encoder_outputs.past_key_values,
+            hidden_states=encoder_outputs.hidden_states,
+            attentions=encoder_outputs.attentions,
+            cross_attentions=encoder_outputs.cross_attentions,
+        )
+
+
+
+class BertLMHeadModel(BertPreTrainedModel):
+
+    _keys_to_ignore_on_load_unexpected = [r"pooler"]
+    _keys_to_ignore_on_load_missing = [r"position_ids", r"predictions.decoder.bias"]
+
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.bert = BertModel(config, add_pooling_layer=False)
+        self.cls = BertOnlyMLMHead(config)
+
+        self.init_weights()
+
+    def get_output_embeddings(self):
+        return self.cls.predictions.decoder
+
+    def set_output_embeddings(self, new_embeddings):
+        self.cls.predictions.decoder = new_embeddings
+
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        labels=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+        return_logits=False,            
+        is_decoder=True,
+        reduction='mean',
+        mode='multimodal', 
+    ):
+        r"""
+        encoder_hidden_states  (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
+            the model is configured as a decoder.
+        encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
+            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
+            the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
+            Labels for computing the left-to-right language modeling loss (next word prediction). Indices should be in
+            ``[-100, 0, ..., config.vocab_size]`` (see ``input_ids`` docstring) Tokens with indices set to ``-100`` are
+            ignored (masked), the loss is only computed for the tokens with labels n ``[0, ..., config.vocab_size]``
+        past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
+            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
+            If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
+            (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
+            instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
+        use_cache (:obj:`bool`, `optional`):
+            If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
+            decoding (see :obj:`past_key_values`).
+        Returns:
+        Example::
+            >>> from transformers import BertTokenizer, BertLMHeadModel, BertConfig
+            >>> import torch
+            >>> tokenizer = BertTokenizer.from_pretrained('bert-base-cased')
+            >>> config = BertConfig.from_pretrained("bert-base-cased")
+            >>> model = BertLMHeadModel.from_pretrained('bert-base-cased', config=config)
+            >>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt")
+            >>> outputs = model(**inputs)
+            >>> prediction_logits = outputs.logits
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        if labels is not None:
+            use_cache = False
+
+        outputs = self.bert(
+            input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            is_decoder=is_decoder,
+            mode=mode,
+        )
+        
+        sequence_output = outputs[0]
+        prediction_scores = self.cls(sequence_output)
+        
+        if return_logits:
+            return prediction_scores[:, :-1, :].contiguous()  
+
+        lm_loss = None
+        if labels is not None:
+            # we are doing next-token prediction; shift prediction scores and input ids by one
+            shifted_prediction_scores = prediction_scores[:, :-1, :].contiguous()
+            labels = labels[:, 1:].contiguous()
+            loss_fct = CrossEntropyLoss(reduction=reduction, label_smoothing=0.1) 
+            lm_loss = loss_fct(shifted_prediction_scores.view(-1, self.config.vocab_size), labels.view(-1))
+            if reduction=='none':
+                lm_loss = lm_loss.view(prediction_scores.size(0),-1).sum(1)               
+
+        if not return_dict:
+            output = (prediction_scores,) + outputs[2:]
+            return ((lm_loss,) + output) if lm_loss is not None else output
+
+        return CausalLMOutputWithCrossAttentions(
+            loss=lm_loss,
+            logits=prediction_scores,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            cross_attentions=outputs.cross_attentions,
+        )
+
+    def prepare_inputs_for_generation(self, input_ids, past=None, attention_mask=None, **model_kwargs):
+        input_shape = input_ids.shape
+        # if model is used as a decoder in encoder-decoder model, the decoder attention mask is created on the fly
+        if attention_mask is None:
+            attention_mask = input_ids.new_ones(input_shape)
+
+        # cut decoder_input_ids if past is used
+        if past is not None:
+            input_ids = input_ids[:, -1:]
+
+        return {
+            "input_ids": input_ids, 
+            "attention_mask": attention_mask, 
+            "past_key_values": past,
+            "encoder_hidden_states": model_kwargs.get("encoder_hidden_states", None),
+            "encoder_attention_mask": model_kwargs.get("encoder_attention_mask", None),
+            "is_decoder": True,
+        }
+
+    def _reorder_cache(self, past, beam_idx):
+        reordered_past = ()
+        for layer_past in past:
+            reordered_past += (tuple(past_state.index_select(0, beam_idx) for past_state in layer_past),)
+        return reordered_past

diffsynth/extensions/ImageQualityMetric/BLIP/vit.py

@@ -0,0 +1,301 @@
+'''
+ * Adapted from BLIP (https://github.com/salesforce/BLIP)
+ * Based on timm code base
+ * https://github.com/rwightman/pytorch-image-models/tree/master/timm
+'''
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from functools import partial
+
+from timm.models.vision_transformer import _cfg, PatchEmbed
+from timm.models.registry import register_model
+from timm.models.layers import trunc_normal_, DropPath
+from timm.models.helpers import named_apply, adapt_input_conv
+
+# from fairscale.nn.checkpoint.checkpoint_activations import checkpoint_wrapper
+
+class Mlp(nn.Module):
+    """ MLP as used in Vision Transformer, MLP-Mixer and related networks
+    """
+    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.act = act_layer()
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.drop = nn.Dropout(drop)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.drop(x)
+        x = self.fc2(x)
+        x = self.drop(x)
+        return x
+
+
+class Attention(nn.Module):
+    def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.):
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        # NOTE scale factor was wrong in my original version, can set manually to be compat with prev weights
+        self.scale = qk_scale or head_dim ** -0.5
+        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Linear(dim, dim)
+        self.proj_drop = nn.Dropout(proj_drop)
+        self.attn_gradients = None
+        self.attention_map = None
+        
+    def save_attn_gradients(self, attn_gradients):
+        self.attn_gradients = attn_gradients
+        
+    def get_attn_gradients(self):
+        return self.attn_gradients
+    
+    def save_attention_map(self, attention_map):
+        self.attention_map = attention_map
+        
+    def get_attention_map(self):
+        return self.attention_map
+    
+    def forward(self, x, register_hook=False):
+        B, N, C = x.shape
+        qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
+        q, k, v = qkv[0], qkv[1], qkv[2]   # make torchscript happy (cannot use tensor as tuple)
+
+        attn = (q @ k.transpose(-2, -1)) * self.scale
+        attn = attn.softmax(dim=-1)
+        attn = self.attn_drop(attn)
+                
+        if register_hook:
+            self.save_attention_map(attn)
+            attn.register_hook(self.save_attn_gradients)        
+
+        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+
+class Block(nn.Module):
+
+    def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
+                 drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm, use_grad_checkpointing=False):
+        super().__init__()
+        self.norm1 = norm_layer(dim)
+        self.attn = Attention(
+            dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
+        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
+        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
+        self.norm2 = norm_layer(dim)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
+
+        # if use_grad_checkpointing:
+        #     self.attn = checkpoint_wrapper(self.attn)
+        #     self.mlp = checkpoint_wrapper(self.mlp)
+
+    def forward(self, x, register_hook=False):
+        x = x + self.drop_path(self.attn(self.norm1(x), register_hook=register_hook))
+        x = x + self.drop_path(self.mlp(self.norm2(x)))
+        return x
+
+    
+class VisionTransformer(nn.Module):
+    """ Vision Transformer
+    A PyTorch impl of : `An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale`  -
+        https://arxiv.org/abs/2010.11929
+    """
+    def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, embed_dim=768, depth=12,
+                 num_heads=12, mlp_ratio=4., qkv_bias=True, qk_scale=None, representation_size=None,
+                 drop_rate=0., attn_drop_rate=0., drop_path_rate=0., norm_layer=None, 
+                 use_grad_checkpointing=False, ckpt_layer=0):
+        """
+        Args:
+            img_size (int, tuple): input image size
+            patch_size (int, tuple): patch size
+            in_chans (int): number of input channels
+            num_classes (int): number of classes for classification head
+            embed_dim (int): embedding dimension
+            depth (int): depth of transformer
+            num_heads (int): number of attention heads
+            mlp_ratio (int): ratio of mlp hidden dim to embedding dim
+            qkv_bias (bool): enable bias for qkv if True
+            qk_scale (float): override default qk scale of head_dim ** -0.5 if set
+            representation_size (Optional[int]): enable and set representation layer (pre-logits) to this value if set
+            drop_rate (float): dropout rate
+            attn_drop_rate (float): attention dropout rate
+            drop_path_rate (float): stochastic depth rate
+            norm_layer: (nn.Module): normalization layer
+        """
+        super().__init__()
+        self.num_features = self.embed_dim = embed_dim  # num_features for consistency with other models
+        norm_layer = norm_layer or partial(nn.LayerNorm, eps=1e-6)
+
+        self.patch_embed = PatchEmbed(
+            img_size=img_size, patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim)
+
+        num_patches = self.patch_embed.num_patches
+
+        self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
+        self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim))
+        self.pos_drop = nn.Dropout(p=drop_rate)
+
+        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)]  # stochastic depth decay rule
+        self.blocks = nn.ModuleList([
+            Block(
+                dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, qk_scale=qk_scale,
+                drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer,
+                use_grad_checkpointing=(use_grad_checkpointing and i>=depth-ckpt_layer)
+            )
+            for i in range(depth)])
+        self.norm = norm_layer(embed_dim)
+
+        trunc_normal_(self.pos_embed, std=.02)
+        trunc_normal_(self.cls_token, std=.02)
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight, std=.02)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                nn.init.constant_(m.bias, 0)
+        elif isinstance(m, nn.LayerNorm):
+            nn.init.constant_(m.bias, 0)
+            nn.init.constant_(m.weight, 1.0)
+
+    @torch.jit.ignore
+    def no_weight_decay(self):
+        return {'pos_embed', 'cls_token'}
+
+    def forward(self, x, register_blk=-1):
+        B = x.shape[0]
+        x = self.patch_embed(x)
+
+        cls_tokens = self.cls_token.expand(B, -1, -1)  # stole cls_tokens impl from Phil Wang, thanks
+        x = torch.cat((cls_tokens, x), dim=1)
+  
+        x = x + self.pos_embed[:,:x.size(1),:]
+        x = self.pos_drop(x)
+
+        for i,blk in enumerate(self.blocks):
+            x = blk(x, register_blk==i)
+        x = self.norm(x)
+        
+        return x
+
+    @torch.jit.ignore()
+    def load_pretrained(self, checkpoint_path, prefix=''):
+        _load_weights(self, checkpoint_path, prefix)
+        
+
+@torch.no_grad()
+def _load_weights(model: VisionTransformer, checkpoint_path: str, prefix: str = ''):
+    """ Load weights from .npz checkpoints for official Google Brain Flax implementation
+    """
+    import numpy as np
+
+    def _n2p(w, t=True):
+        if w.ndim == 4 and w.shape[0] == w.shape[1] == w.shape[2] == 1:
+            w = w.flatten()
+        if t:
+            if w.ndim == 4:
+                w = w.transpose([3, 2, 0, 1])
+            elif w.ndim == 3:
+                w = w.transpose([2, 0, 1])
+            elif w.ndim == 2:
+                w = w.transpose([1, 0])
+        return torch.from_numpy(w)
+
+    w = np.load(checkpoint_path)
+    if not prefix and 'opt/target/embedding/kernel' in w:
+        prefix = 'opt/target/'
+
+    if hasattr(model.patch_embed, 'backbone'):
+        # hybrid
+        backbone = model.patch_embed.backbone
+        stem_only = not hasattr(backbone, 'stem')
+        stem = backbone if stem_only else backbone.stem
+        stem.conv.weight.copy_(adapt_input_conv(stem.conv.weight.shape[1], _n2p(w[f'{prefix}conv_root/kernel'])))
+        stem.norm.weight.copy_(_n2p(w[f'{prefix}gn_root/scale']))
+        stem.norm.bias.copy_(_n2p(w[f'{prefix}gn_root/bias']))
+        if not stem_only:
+            for i, stage in enumerate(backbone.stages):
+                for j, block in enumerate(stage.blocks):
+                    bp = f'{prefix}block{i + 1}/unit{j + 1}/'
+                    for r in range(3):
+                        getattr(block, f'conv{r + 1}').weight.copy_(_n2p(w[f'{bp}conv{r + 1}/kernel']))
+                        getattr(block, f'norm{r + 1}').weight.copy_(_n2p(w[f'{bp}gn{r + 1}/scale']))
+                        getattr(block, f'norm{r + 1}').bias.copy_(_n2p(w[f'{bp}gn{r + 1}/bias']))
+                    if block.downsample is not None:
+                        block.downsample.conv.weight.copy_(_n2p(w[f'{bp}conv_proj/kernel']))
+                        block.downsample.norm.weight.copy_(_n2p(w[f'{bp}gn_proj/scale']))
+                        block.downsample.norm.bias.copy_(_n2p(w[f'{bp}gn_proj/bias']))
+        embed_conv_w = _n2p(w[f'{prefix}embedding/kernel'])
+    else:
+        embed_conv_w = adapt_input_conv(
+            model.patch_embed.proj.weight.shape[1], _n2p(w[f'{prefix}embedding/kernel']))
+    model.patch_embed.proj.weight.copy_(embed_conv_w)
+    model.patch_embed.proj.bias.copy_(_n2p(w[f'{prefix}embedding/bias']))
+    model.cls_token.copy_(_n2p(w[f'{prefix}cls'], t=False))
+    pos_embed_w = _n2p(w[f'{prefix}Transformer/posembed_input/pos_embedding'], t=False)
+    if pos_embed_w.shape != model.pos_embed.shape:
+        pos_embed_w = resize_pos_embed(  # resize pos embedding when different size from pretrained weights
+            pos_embed_w, model.pos_embed, getattr(model, 'num_tokens', 1), model.patch_embed.grid_size)
+    model.pos_embed.copy_(pos_embed_w)
+    model.norm.weight.copy_(_n2p(w[f'{prefix}Transformer/encoder_norm/scale']))
+    model.norm.bias.copy_(_n2p(w[f'{prefix}Transformer/encoder_norm/bias']))
+#     if isinstance(model.head, nn.Linear) and model.head.bias.shape[0] == w[f'{prefix}head/bias'].shape[-1]:
+#         model.head.weight.copy_(_n2p(w[f'{prefix}head/kernel']))
+#         model.head.bias.copy_(_n2p(w[f'{prefix}head/bias']))
+#     if isinstance(getattr(model.pre_logits, 'fc', None), nn.Linear) and f'{prefix}pre_logits/bias' in w:
+#         model.pre_logits.fc.weight.copy_(_n2p(w[f'{prefix}pre_logits/kernel']))
+#         model.pre_logits.fc.bias.copy_(_n2p(w[f'{prefix}pre_logits/bias']))
+    for i, block in enumerate(model.blocks.children()):
+        block_prefix = f'{prefix}Transformer/encoderblock_{i}/'
+        mha_prefix = block_prefix + 'MultiHeadDotProductAttention_1/'
+        block.norm1.weight.copy_(_n2p(w[f'{block_prefix}LayerNorm_0/scale']))
+        block.norm1.bias.copy_(_n2p(w[f'{block_prefix}LayerNorm_0/bias']))
+        block.attn.qkv.weight.copy_(torch.cat([
+            _n2p(w[f'{mha_prefix}{n}/kernel'], t=False).flatten(1).T for n in ('query', 'key', 'value')]))
+        block.attn.qkv.bias.copy_(torch.cat([
+            _n2p(w[f'{mha_prefix}{n}/bias'], t=False).reshape(-1) for n in ('query', 'key', 'value')]))
+        block.attn.proj.weight.copy_(_n2p(w[f'{mha_prefix}out/kernel']).flatten(1))
+        block.attn.proj.bias.copy_(_n2p(w[f'{mha_prefix}out/bias']))
+        for r in range(2):
+            getattr(block.mlp, f'fc{r + 1}').weight.copy_(_n2p(w[f'{block_prefix}MlpBlock_3/Dense_{r}/kernel']))
+            getattr(block.mlp, f'fc{r + 1}').bias.copy_(_n2p(w[f'{block_prefix}MlpBlock_3/Dense_{r}/bias']))
+        block.norm2.weight.copy_(_n2p(w[f'{block_prefix}LayerNorm_2/scale']))
+        block.norm2.bias.copy_(_n2p(w[f'{block_prefix}LayerNorm_2/bias']))
+
+            
+def interpolate_pos_embed(pos_embed_checkpoint, visual_encoder):        
+    # interpolate position embedding
+    embedding_size = pos_embed_checkpoint.shape[-1]
+    num_patches = visual_encoder.patch_embed.num_patches
+    num_extra_tokens = visual_encoder.pos_embed.shape[-2] - num_patches
+    # height (== width) for the checkpoint position embedding
+    orig_size = int((pos_embed_checkpoint.shape[-2] - num_extra_tokens) ** 0.5)
+    # height (== width) for the new position embedding
+    new_size = int(num_patches ** 0.5)
+
+    if orig_size!=new_size:
+        # class_token and dist_token are kept unchanged
+        extra_tokens = pos_embed_checkpoint[:, :num_extra_tokens]
+        # only the position tokens are interpolated
+        pos_tokens = pos_embed_checkpoint[:, num_extra_tokens:]
+        pos_tokens = pos_tokens.reshape(-1, orig_size, orig_size, embedding_size).permute(0, 3, 1, 2)
+        pos_tokens = torch.nn.functional.interpolate(
+            pos_tokens, size=(new_size, new_size), mode='bicubic', align_corners=False)
+        pos_tokens = pos_tokens.permute(0, 2, 3, 1).flatten(1, 2)
+        new_pos_embed = torch.cat((extra_tokens, pos_tokens), dim=1)
+        print('reshape position embedding from %d to %d'%(orig_size ** 2,new_size ** 2))
+        
+        return new_pos_embed    
+    else:
+        return pos_embed_checkpoint

diffsynth/extensions/ImageQualityMetric/__init__.py

@@ -0,0 +1,148 @@
+from modelscope import snapshot_download
+from typing_extensions import Literal, TypeAlias
+import os
+from diffsynth.extensions.ImageQualityMetric.aesthetic import AestheticScore
+from diffsynth.extensions.ImageQualityMetric.imagereward import ImageRewardScore
+from diffsynth.extensions.ImageQualityMetric.pickscore import PickScore
+from diffsynth.extensions.ImageQualityMetric.clip import CLIPScore
+from diffsynth.extensions.ImageQualityMetric.hps import HPScore_v2
+from diffsynth.extensions.ImageQualityMetric.mps import MPScore
+
+
+preference_model_id: TypeAlias = Literal[
+    "ImageReward",
+    "Aesthetic",
+    "PickScore",
+    "CLIP",
+    "HPSv2",
+    "HPSv2.1",
+    "MPS",
+]
+model_dict = {
+    "ImageReward": {
+        "model_id": "DiffSynth-Studio/QualityMetric_reward_pretrained",
+        "allow_file_pattern": [
+            "ImageReward/ImageReward.safetensors",
+            "ImageReward/med_config.json",
+            "bert-base-uncased/config.json",
+            "bert-base-uncased/model.safetensors",
+            "bert-base-uncased/tokenizer.json",
+            "bert-base-uncased/tokenizer_config.json",
+            "bert-base-uncased/vocab.txt",
+        ],
+        "load_path": {
+            "imagereward": "ImageReward/ImageReward.safetensors",
+            "med_config": "ImageReward/med_config.json",
+            "bert_model_path": "bert-base-uncased",
+        },
+        "model_class": ImageRewardScore
+    },
+    "Aesthetic": {
+        "model_id": "DiffSynth-Studio/QualityMetric_reward_pretrained",
+        "allow_file_pattern": [
+            "aesthetic-predictor/sac+logos+ava1-l14-linearMSE.safetensors",
+            "clip-vit-large-patch14/config.json",
+            "clip-vit-large-patch14/merges.txt",
+            "clip-vit-large-patch14/model.safetensors",
+            "clip-vit-large-patch14/preprocessor_config.json",
+            "clip-vit-large-patch14/special_tokens_map.json",
+            "clip-vit-large-patch14/tokenizer.json",
+            "clip-vit-large-patch14/tokenizer_config.json",
+            "clip-vit-large-patch14/vocab.json",
+        ],
+        "load_path": {
+            "aesthetic_predictor": "aesthetic-predictor/sac+logos+ava1-l14-linearMSE.safetensors",
+            "clip-large": "clip-vit-large-patch14",
+        },
+        "model_class": AestheticScore
+    },
+    "PickScore": {
+        "model_id": "DiffSynth-Studio/QualityMetric_reward_pretrained",
+        "allow_file_pattern": [
+            "PickScore_v1/*",
+            "CLIP-ViT-H-14-laion2B-s32B-b79K/config.json",
+            "CLIP-ViT-H-14-laion2B-s32B-b79K/merges.txt",
+            "CLIP-ViT-H-14-laion2B-s32B-b79K/preprocessor_config.json",
+            "CLIP-ViT-H-14-laion2B-s32B-b79K/special_tokens_map.json",
+            "CLIP-ViT-H-14-laion2B-s32B-b79K/tokenizer.json",
+            "CLIP-ViT-H-14-laion2B-s32B-b79K/tokenizer_config.json",
+            "CLIP-ViT-H-14-laion2B-s32B-b79K/vocab.json",
+        ],
+        "load_path": {
+            "pickscore": "PickScore_v1",
+            "clip": "CLIP-ViT-H-14-laion2B-s32B-b79K",
+        },
+        "model_class": PickScore
+    },
+    "CLIP": {
+        "model_id": "DiffSynth-Studio/QualityMetric_reward_pretrained",
+        "allow_file_pattern": [
+            "CLIP-ViT-H-14-laion2B-s32B-b79K/open_clip_pytorch_model.bin",
+            "bpe_simple_vocab_16e6.txt.gz",
+        ],
+        "load_path": {
+            "open_clip": "CLIP-ViT-H-14-laion2B-s32B-b79K/open_clip_pytorch_model.bin",
+            "open_clip_bpe": "bpe_simple_vocab_16e6.txt.gz",
+        },
+        "model_class": CLIPScore
+    },
+    "HPSv2": {
+        "model_id": "DiffSynth-Studio/QualityMetric_reward_pretrained",
+        "allow_file_pattern": [
+            "HPS_v2/HPS_v2_compressed.safetensors",
+            "bpe_simple_vocab_16e6.txt.gz",
+        ],
+        "load_path": {
+            "hpsv2": "HPS_v2/HPS_v2_compressed.safetensors",
+            "open_clip_bpe": "bpe_simple_vocab_16e6.txt.gz",
+        },
+        "model_class": HPScore_v2,
+        "extra_kwargs": {"model_version": "v2"}
+    },
+    "HPSv2.1": {
+        "model_id": "DiffSynth-Studio/QualityMetric_reward_pretrained",
+        "allow_file_pattern": [
+            "HPS_v2/HPS_v2.1_compressed.safetensors",
+            "bpe_simple_vocab_16e6.txt.gz",
+        ],
+        "load_path": {
+            "hpsv2.1": "HPS_v2/HPS_v2.1_compressed.safetensors",
+            "open_clip_bpe": "bpe_simple_vocab_16e6.txt.gz",
+        },
+        "model_class": HPScore_v2,
+        "extra_kwargs": {"model_version": "v21"}
+    },
+    "MPS": {
+        "model_id": "DiffSynth-Studio/QualityMetric_reward_pretrained",
+        "allow_file_pattern": [
+            "MPS_overall_checkpoint/MPS_overall_checkpoint_diffsynth.safetensors",
+            "CLIP-ViT-H-14-laion2B-s32B-b79K/config.json",
+            "CLIP-ViT-H-14-laion2B-s32B-b79K/merges.txt",
+            "CLIP-ViT-H-14-laion2B-s32B-b79K/preprocessor_config.json",
+            "CLIP-ViT-H-14-laion2B-s32B-b79K/special_tokens_map.json",
+            "CLIP-ViT-H-14-laion2B-s32B-b79K/tokenizer.json",
+            "CLIP-ViT-H-14-laion2B-s32B-b79K/tokenizer_config.json",
+            "CLIP-ViT-H-14-laion2B-s32B-b79K/vocab.json",
+        ],
+        "load_path": {
+            "mps": "MPS_overall_checkpoint/MPS_overall_checkpoint_diffsynth.safetensors",
+            "clip": "CLIP-ViT-H-14-laion2B-s32B-b79K",
+        },
+        "model_class": MPScore
+    },
+}
+
+
+def download_preference_model(model_name: preference_model_id, cache_dir="models"):
+    metadata = model_dict[model_name]
+    snapshot_download(model_id=metadata["model_id"], allow_file_pattern=metadata["allow_file_pattern"], cache_dir=cache_dir)
+    load_path = metadata["load_path"]
+    load_path = {key: os.path.join(cache_dir, metadata["model_id"], path) for key, path in load_path.items()}
+    return load_path
+
+
+def load_preference_model(model_name: preference_model_id, device = "cuda", path = None):
+    model_class = model_dict[model_name]["model_class"]
+    extra_kwargs = model_dict[model_name].get("extra_kwargs", {})
+    preference_model = model_class(device=device, path=path, **extra_kwargs)
+    return preference_model

diffsynth/extensions/ImageQualityMetric/aesthetic.py

@@ -0,0 +1,148 @@
+from typing import List, Optional
+from PIL import Image
+import torch
+from transformers import AutoProcessor, AutoModel
+from safetensors.torch import load_file
+import os
+from typing import Union, List
+from .config import MODEL_PATHS
+
+class MLP(torch.nn.Module):
+    def __init__(self, input_size: int, xcol: str = "emb", ycol: str = "avg_rating"):
+        super().__init__()
+        self.input_size = input_size
+        self.xcol = xcol
+        self.ycol = ycol
+        self.layers = torch.nn.Sequential(
+            torch.nn.Linear(self.input_size, 1024),
+            #torch.nn.ReLU(),
+            torch.nn.Dropout(0.2),
+            torch.nn.Linear(1024, 128),
+            #torch.nn.ReLU(),
+            torch.nn.Dropout(0.2),
+            torch.nn.Linear(128, 64),
+            #torch.nn.ReLU(),
+            torch.nn.Dropout(0.1),
+            torch.nn.Linear(64, 16),
+            #torch.nn.ReLU(),
+            torch.nn.Linear(16, 1),
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return self.layers(x)
+
+    def training_step(self, batch: dict, batch_idx: int) -> torch.Tensor:
+        x = batch[self.xcol]
+        y = batch[self.ycol].reshape(-1, 1)
+        x_hat = self.layers(x)
+        loss = torch.nn.functional.mse_loss(x_hat, y)
+        return loss
+
+    def validation_step(self, batch: dict, batch_idx: int) -> torch.Tensor:
+        x = batch[self.xcol]
+        y = batch[self.ycol].reshape(-1, 1)
+        x_hat = self.layers(x)
+        loss = torch.nn.functional.mse_loss(x_hat, y)
+        return loss
+
+    def configure_optimizers(self) -> torch.optim.Optimizer:
+        return torch.optim.Adam(self.parameters(), lr=1e-3)
+
+
+class AestheticScore(torch.nn.Module):
+    def __init__(self, device: torch.device, path: str = MODEL_PATHS):
+        super().__init__()
+        self.device = device
+        self.aes_model_path = path.get("aesthetic_predictor")
+        # Load the MLP model
+        self.model = MLP(768)
+        try:
+            if self.aes_model_path.endswith(".safetensors"):
+                state_dict = load_file(self.aes_model_path)
+            else:
+                state_dict = torch.load(self.aes_model_path)
+            self.model.load_state_dict(state_dict)
+        except Exception as e:
+            raise ValueError(f"Error loading model weights from {self.aes_model_path}: {e}")
+
+        self.model.to(device)
+        self.model.eval()
+
+        # Load the CLIP model and processor
+        clip_model_name = path.get('clip-large')
+        self.model2 = AutoModel.from_pretrained(clip_model_name).eval().to(device)
+        self.processor = AutoProcessor.from_pretrained(clip_model_name)
+
+    def _calculate_score(self, image: torch.Tensor) -> float:
+        """Calculate the aesthetic score for a single image.
+
+        Args:
+            image (torch.Tensor): The processed image tensor.
+
+        Returns:
+            float: The aesthetic score.
+        """
+        with torch.no_grad():
+            # Get image embeddings
+            image_embs = self.model2.get_image_features(image)
+            image_embs = image_embs / torch.norm(image_embs, dim=-1, keepdim=True)
+
+            # Compute score
+            score = self.model(image_embs).cpu().flatten().item()
+
+        return score
+
+    @torch.no_grad()
+    def score(self, images: Union[str, List[str], Image.Image, List[Image.Image]], prompt: str = "") -> List[float]:
+        """Score the images based on their aesthetic quality.
+
+        Args:
+            images (Union[str, List[str], Image.Image, List[Image.Image]]): Path(s) to the image(s) or PIL image(s).
+
+        Returns:
+            List[float]: List of scores for the images.
+        """
+        try:
+            if isinstance(images, (str, Image.Image)):
+                # Single image
+                if isinstance(images, str):
+                    pil_image = Image.open(images)
+                else:
+                    pil_image = images
+                
+                # Prepare image inputs
+                image_inputs = self.processor(
+                    images=pil_image,
+                    padding=True,
+                    truncation=True,
+                    max_length=77,
+                    return_tensors="pt",
+                ).to(self.device)
+
+                return [self._calculate_score(image_inputs["pixel_values"])]
+            elif isinstance(images, list):
+                # Multiple images
+                scores = []
+                for one_image in images:
+                    if isinstance(one_image, str):
+                        pil_image = Image.open(one_image)
+                    elif isinstance(one_image, Image.Image):
+                        pil_image = one_image
+                    else:
+                        raise TypeError("The type of parameter images is illegal.")
+                    
+                    # Prepare image inputs
+                    image_inputs = self.processor(
+                        images=pil_image,
+                        padding=True,
+                        truncation=True,
+                        max_length=77,
+                        return_tensors="pt",
+                    ).to(self.device)
+
+                    scores.append(self._calculate_score(image_inputs["pixel_values"]))
+                return scores
+            else:
+                raise TypeError("The type of parameter images is illegal.")
+        except Exception as e:
+            raise RuntimeError(f"Error in scoring images: {e}")