Description:
This workflow allows you to generate images from another images. It is optimized to integrate all metadata for simplified online publishing.
Resources you need:
📂Files :
For base version
Model : fp8 or bf16
in models\diffusion_models
For GGUF version
Model : Q8,Q6,Q5,Q4,Q3
in models\unet
VAE : qwen_image_vae.safetensors
in models\vae
CLIP : qwen_2.5_vl_7b_fp8_scaled.safetensors
in models\clip
LoRA : 8steps, 4steps
in models\loras
📦Custom Nodes :
Description
base version
FAQ
Comments (8)
When using qwen_image_edit_fp8_e4m3fn.safetensors
You need to turn off --fast(FP16 accumulation) argument
This argument causes image corruption
gguf is fine
turn off --fast(FP16 accumulation) argument - where?
@MugenMan bat file that runs Comfy (run_gpu_nvidia.bat)
@tmxpdlqmfelvbwjs I don't know, I also have run_nvidia_gpu.bat, which doesn't have these arguments. I always ran it, and run_nvidia_gpu_fast_fp16_accumulation.bat, which does have these arguments
Anyway, if i set it to default in the node, it also works
@MugenMan Yes, default is off
Now Qwen image edit2509 is better
Workflow updated
can it make very nsfw?