saddam213/stable-diffusion-3.5-lite-onnx

Stable-Diffusion 3.5 Lite - Onnx Olive DirectML Optimized

Information:

This conversion uses int4 data type for the TextEncoder3, this drops VRAM requirement to between 8GB and 16GB However this will result in a slight quality drop compared to the base stable-diffusion-3.5-medium model

Original Model

https://huggingface.co/stabilityai/stable-diffusion-3.5-medium

C# Inference Demo

https://github.com/TensorStack-AI/OnnxStack

// Create Pipeline
var pipeline = StableDiffusion3Pipeline.CreatePipeline("D:\\Models\\stable-diffusion-3.5-lite-onnx");

// Prompt
var promptOptions = new PromptOptions
{
    Prompt = "Create a scene of a mystical shaman, with animal skins and feathers, performing a ritual in a sacred grove."
};

// Run pipeline
var result = await pipeline.GenerateImageAsync(promptOptions);

// Save Image Result
await result.SaveAsync("Result.png");

Inference Result