PLEASE RELEASE THE COMFY UI WORKFLOW!!!!!!
i need to use this lora with qwen image 2512 in order to speed up the generation.
I have a template called "Qwen-Image 2512 Turbo" in ComfyUI.
Update to the latest version (I do a git pull on master).
Download Wuli-Qwen-Image-2512-Turbo-LoRA-2steps-V1.0-bf16.safetensors in models/loras
I have a template called "Qwen-Image 2512 Turbo" in ComfyUI.
Update to the latest version (I do a git pull on master).
Download Wuli-Qwen-Image-2512-Turbo-LoRA-2steps-V1.0-bf16.safetensors in models/loras
Hi! Tried that but I get some kind of ugly results which I want to share here.
My PNG should include JSON for ComfyUI. I used the template "as is" with an example prompt from here. Not nearly the quality shown on the HF page sadly...
THX for having a look into it!
EDIT: I've already found the solution! The problem is caused by the FP8 model used in the ComfyUI template! If you use the full BF16 model from QWEN Image 2512, you get full image quality!
GGUF works just as well as BF16, since it is internally decoded to BF16.