zh-CN/tutorials/image/qwen/qwen-image-edit

[giscus[bot]]

tutorials/image/qwen/qwen-image-edit

Qwen-Image-Edit 是 Qwen-Image 的图像编辑版本，基于20B模型进一步训练，支持精准文字编辑和语义/外观双重编辑能力。

https://docs.comfy.org/tutorials/image/qwen/qwen-image-edit

[mald55]

When will the dekstop version of comfyui be updated to support this model?

[comfyui-wiki]

Now, the stable version is released, so the Desktop supports this model now.

[Armanovich]

Thank you for the workflow, but you should mention that it will only produce black images if safe attention is enabled — you need to disable safe attention for it to work.

[Armanovich]

I mean sage attention, of corse

[HenkDz]

thanks man, was wondering for sometime.

[Artm1st]

Hey! I just used a patch Sage Attention KJ Node to disable sage_attention but still have that black image, how did you disabled it?

[completeai]

Wondering the same Art

[completeai]

Getting, - Custom validation failed for node: image - Invalid image file: Qwen-Image_00043_.png

[George5121]

Hello! Big work here, thanks!

i have some problem, can u help please?

"TextEncodeQwenImageEdit
einsum(): subscript j has size 480 for operand 1 which does not broadcast with previously seen size 1440"

[quadjojo]

same issue here

[comfyui-wiki]

Does this issue still exist after updating to the latest ComfyUI?

[George5121]

The problem is gone, thank you very much for your work!

[9Somboon]

Some nodes require a newer version of ComfyUI (current: 0.3.50). Please update to use all nodes.
Requires ComfyUI 0.3.50:
TextEncodeQwenImageEdit

I have updated to the latest version but it still doesn't work.

[Aleksandr-3D]

try updating via update_comfyui.bat in the update folder

[airev9498]

I am having the same issue. were you able to solve this?

[Armanovich]

Does anyone have a better workflow? I’ve tried this one with fp8 and GGUF Q8, and honestly the output is terrible — faces don’t stay consistent, scales and colors shift, and I can’t get any readable text in the images. I tried both English and French, and not a single word came out correctly. Nothing works as it should. For me, FLUX Kontext outperforms Qwen Image Edit by far. But the results can’t be this bad by design, so it must be the workflow… Do you experience the same issues?

[comfyui-wiki]

It's not a workflow issue. Comfy had a few more updates yesterday. Try updating ComfyUI to the latest version and see if it works better now.

[maxidesu]

With almost any input image I use, editing tasks like adding sunglasses or changing clothes make the character shift slightly, with a tendency to zoom in and stretch the image. Sometimes it even alters the colors and quality a bit, whether Scale Image is on or off, and with or without Lighting. That aside, which I’m sure someone will figure out how to fix, it does an incredible job at editing.

[comfyui-wiki]

Try update ComfyUI to 0.3.51

[cheese7828]

comfyui 穩定版 什麼時候 會更新Qwen image edit ???等好久

[comfyui-wiki]

目前已经更新了

[Aries1989]

I want a workflow which could edit image. 🪶

[completeai]

You can, upload your image. As a photographer this is GREAT

[iamdagy]

‪Great video. But how did you set it up to run in a rtx5090 . Im trying but I’m getting out of memory . I have to either disable the 4 steps Lora’s or offload the clip to the cpu‬

[shadowstar4222]

For section 3, what's the difference between the green and blue text encoder? Is one a negative and one a positive?

[NotThePassword]

Also wondering this.

[Artm1st]

Hey! I don't why but my outputs are black empty image, It runs all the WF but the result is just a black canvas, I have tried with the example input, and the result is the same, a black canvas.

[George5121]

Check if your prompt has text inside brackets "()". For some reason, the presence of these brackets caused a black square for me.

[YouYouTheBoxx]

Very simple it seems: try to disable sage atyention: it doesn't work with it.

[francovp]

what are the hardware requirements?

[Callisin]

Not working on Mac M2 ?

[jonny7737]

Mostly worked on my M2 MAX 64GB.

Prompt executed in 13:43 - But this is a mac so happy it worked.

[AntZ7Z]

Is it possible to add a second image to this workflow so that images can be combined, objects from one image can be replaced with objects from another image, etc.?

[prankousky]

I was wondering the same. It'd be so cool to add one reference (style) image, then have qwen edit some source image (photo). "transfer the style from reference to original, remove background, make subject wear a tuxedo and top hat".

Style transfer works without reference images as well, but not all styles are available, so adding a second image would be a real enhancement :)

[guyouyue]

我采用默认的工作流json和对应的图片生成出来的结果是一片雪花块是怎么回事?

[comfyui-wiki]

有更新吗？ 另外如果用的是秋叶的整合包的话需要把激进模式关掉

[completeai]

I cant download the json file

[completeai]

Nevermind! It worked now.

[Suif-Vayne]

Need multi-image workflow support!

[unhappyface]

Hi recommended graphic card for this model?

[githubenj]

thank you, Comfy lords!

[hkechung]

I am working on Mac m2 with 32gb ram
When I ran it got Sampler error " Trying to convert Float8_e4m3fn to the MPS backend but it does not have support for that dtype.

[hkechung]

I switch to M1 Pro with 64GB ram. It works now.

[Vadim-Podlevsky]

Running any of the 3 suggested workflow configurations with a test file from this article in comfyui web version 0.3.57 and python 3.13 it takes an enormous (89.53gb to be exact) amount of unified memory on M1 Max 64 GB, digging deep into a swap file and sucking all the resources from OS. The process takes forever until I shut id down manually, and the last active node is KSampler. Any ideas?

[matrix12315]

I am a newbie, could anybody explain why the files are different from official? I only want to download once for both inference and fine-tune. It seems I couldnt fine-tune if I download the files above.