Workflow
#1
by
ryg81
- opened
Hey any workflow on what settings works best?
I used the native ComfyUI workflow here: https://docs.comfy.org/tutorials/image/qwen/qwen-image and adapted it to use all GGUFs loaders to work with less VRAM. Does well and can generate batches of 4 images simultenously on my 3090TI FE 24GB VRAM @ 350Watts (nvidia -smi -pl 350
) in a couple minutes.
Then download these files and place them in the following directories:
- https://huggingface.co/QuantStack/Qwen-Image-Distill-GGUF?show_file_info=Qwen_Image_Distill-Q6_K.gguf saved to
./models/unet/QwenImg/Qwen_Image_Distill-Q6_K.gguf
- https://huggingface.co/unsloth/Qwen2.5-VL-7B-Instruct-GGUF?show_file_info=Qwen2.5-VL-7B-Instruct-UD-Q6_K_XL.gguf saved to
./models/text_encoders/QwenImg/Qwen2.5-VL-7B-Instruct-UD-Q6_K_XL.gguf
(don't tell anyone I use unsloth quants ;p ) - https://huggingface.co/QuantStack/Qwen-Image-GGUF/blob/main/VAE/Qwen_Image-VAE.safetensors saved to
./models/vae/QwenImg/Qwen_Image-VAE.safetensors
Finally with some like might have Qwen-Image LoRAs working in ComfyUI who knows: https://huggingface.co/flymy-ai/qwen-image-realism-lora/discussions/1 haha...
Good luck!
YarvixPA
changed discussion title from
Any workflow?
to Workflow
thanks
ryg81
changed discussion status to
closed