Mathemagic's ComfyUI Workflows [Qwen family]
Catgirl because why not.
This post publishes my Qwen-Image and Qwen-Image-Edit workflows. For other model families, see:
Z-Image Turbo workflows
Chroma1-HD workflows
Illustrious workflows
Tools: background remover workflow
The workflows are embedded in the images. The text on the sign says which workflow is embedded in the image.
For more details, see my guide.
Notes
It took long enough (and then some) to do this. But here it is.
See the text on the signs for which is what.
Qwen couldn't spell some of the words: "img2img" became "img2ag", while "inpaint" invariably became "inpraint". In such cases, I went for clear text descriptions instead.
About this post
The cover image is a Qwen edit of my guide's cover image, automatically outpainting as needed. It was made with the Unsloth Q4_K_M quant of Qwen-Image-Edit 2511, but that hasn't been posted on this site, so I picked the closest existing one.
The image input to the Edit 2511 image was the Qwen 2512 txt2img image.
Which Techniques tags to put on a Qwen edit is a head-scratcher. All other cases are straightforward - it's a workflow, plus the operation that was done by that workflow (txt2img, img2img, inpaint). But is an edit an img2img or an inpaint? It's a kind of very targeted img2img, which often effectively performs an inpaint. I'm leaving the tag as workflow only for now.
Color issues in Qwen-Image-Edit 2511? [solved, see at end]
I've encountered issues with color preservation with Edit 2511. The model easily oversaturates/burns the colors as well as cranks up the brightness in its output - but I haven't yet found the cause.
It could be the VAE if it should use a different one, since I'm using the original Qwen-Image VAE. But I couldn't find another VAE specifically for 2511. So it likely isn't the VAE.
When the accelerator is in use, CFG is always 1, so it isn't that, either.
It could be the Lightning LoRA? Someone on Reddit suggested using the original Lightning LoRA for the original Qwen-Image instead. Tested. Works, but no significant change. The old LoRA has no issues on the model it was designed for, so at least that LoRA itself is probably fine. A priori, it's anyone's guess how well a LoRA trained against the original Qwen-Image applies to Qwen-Image-Edit-2511. But empirically, it works just as fine as the new LoRA that was specifically made for Edit 2511 - with either one, I get the same color issues in Edit 2511.
If 2511 stays uncooperative, it is possible to roll back to 2509. The same workflows work for that version, too.
If you have read this far, and either know of a solution, or can think of something else to try, please ping me in the comments.
EDIT 2026/01/26: This was a ComfyUI version issue. Commit 56fa7dbe380cb5591c5542f8aa51ce2fc26beedf from 7 December 2025 had the color-burning issue, but commit 7ee77ff038937bdfdbea5d603ad8d4c487c14fd6 from 25 January 2026 works fine.

