which qwen_image models does this work with?

#8
by nikkosad - opened

I have downloaded
qwen_image_q4_k_m gguf
this Lora doesn't seem to be working which models have you been working with even the qwen_image_fp8_e4m3fn model is not working
the images all come out either really blurry or low contrast which means i am not using the correct number of steps.

Yes, I found the same problems, when using quantized GGUF qwen-image models, all the 4/8 steps or bf16/non-bf16 loras,
With the correct number of steps, the results are just blur images.

Works perfectly fine with GGUF here.
(and most/many tutorials on youtube for Qwen Image + LightX lora uses GGUF too, so you can even see it work great https://youtu.be/erj5YlR9hvE?t=117 )

Depends on what you mean by blurry though. The Qwen Image has a bit "soft" look, with or without low step lora.s

ok so I found the problem, this seems to a problem with the comfy ui with lora processing, so far the youtube videos are all using the nightly builds so they are all getting it fine. But i am using an installtion of comfyui which is set to stable so its still bugged for me. BUT right now comfy pushed out an update so its working now!! woohooo
comfy.jpg

Sign up or log in to comment