I was trying to create Qwen-Image Workflow ASAP ;)
get GGUF: https://huggingface.co/city96/Qwen-Image-gguf/tree/main
get VAE and CLIP: https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/tree/main/split_files
Update: there are at least two Abliterated (Uncensored to a degree) GGUF versions of CLIP text-encoder for this model:
1) https://huggingface.co/mradermacher/Qwen2.5-VL-7B-Instruct-abliterated-GGUF/tree/main
2) https://huggingface.co/mradermacher/Qwen2.5-VL-7B-Abliterated-Caption-it-GGUF/tree/main <--- this one is my personal favorite!
3) for 4-8 steps lora go to this link: https://huggingface.co/lightx2v/Qwen-Image-Lightning/tree/main
P.S. The model is very sensitive to photography settings. Try to be careful with the depth of field and shallow focus in your prompts.
Description
further improvements and additions to the WF: twisty upscale logic ;)
FAQ
Comments (10)
Thanks for the tip on mradermacher; I've used his models for local llm, but not for a ClipLoader before. I can't really see that there is much difference in image generation though; is there anything specific you need to add to prompt to improve response (not like typical separate system/user prompts I'm used to on llm).
...nice improvement in the advanced
the difference is sometimes subtle - depending on how naughty is your prompt, because when text encoder has nothing to "police" there - then the difference is minor. I've seen really impressive changes even in the most innocent prompts ;)
I found out that better CLIP quantization is the KEY to more image quality. Now using mradermacherer CLIP 16BF gguf 15Gig with K4_0 model.
yes, quality drastically depends on TE
mradermacher/Sensualize-Mixtral-bf16-GGUF - it weighs from 21 to 49 gigabytes, which model do you recommend using, and how much video memory do you have?
I'm unable to use the workflow because there are lots of missing nodes and when I install them with the node manager, I get the message that ComfyUI must be restarted for the installation to complete. After reinstallation, the nodes still are missing and this cycle has no end. I've updated ComfyUI to the most recent nightly version.
the whole point is to add those nodes - without them - just use the default one, but it is limited to those default samplers / schedulers !!!
OliviaRossi Thank you! I managed to solve my problem. In the ComfyUI Manager, the channel was set to Dev rather than Default, for some reason. After switching this back to Default, I was able to find and install the missing nodes.
pierrenormand hmmm, interesting glitch! good to know!