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
this WF uses https://civitai.com/models/2088956 + https://civitai.com/models/2049408 for adding some photorealism ;)
FAQ
Comments (15)
What's is the difference between normal and the abliterated version?
Asking because I'm surprised to see the Q8 been only 8Gb, I think I never hear of it and that was the reason I didn't tried qwen
@0l1v1aR0551 I just downloaded the 8 step workflow, and noticed my issue. I was thinking that the Alliterated was the model, while it is the CLIP only.
@BigDannyPt yup
I'm just now finding your workflow. Can you help me understand what the differences are between the various versions?
use the most fresh one as a starting point, since the "Hi-Res" introduces additional upscale and post-upscale resampling, and it is more complex for understanding due to the nature of ClownShark nodes
@0l1v1aR0551 Thank you
It doesn't look like the middle two samplers do anything? workflow just steps through them? I think one of your previous workflows had a unsample, then resample but both the middle two are resample?
when we have the upscale - then after we have to unsample first, if not - we have to force adding of noise in quants of 5 steps like 3 times, and then do the resampling of the remaining steps in batch, that is why those cycle nodes are placed in that way ---> it is borrowed from the multiple official workflows of the creator of those nodes. And in terminal we can see - it all goes first 3 time x5 steps and then all remaining ones ... Also = the amount of steps in all first cycle nodes can be tweaked for better results of the final output (it required meticulous testing of your prompt vs resolution)
@0l1v1aR0551 thanks for the reply and info. I lurked the sampling madness on L3 discord for so long but never really learned how to get the best from all those nodes. If you watch that channel then you know every day is a different method and new things to test.
@EricRollei21 can you give me a link?
@0l1v1aR0551 Just sent you a DM
@EricRollei21 HUGE TY!!!!
Lora from EPIC WF dose not exist anymore
sad :(