image editing code

[user4042025]

Hi! Thanks for the great work! I was wondering — is there any existing code or support for editing existing images, similar to how text-to-image generation (t2i) is already implemented?

At the moment, I could only find code related to image generation. If editing functionality exists or is planned, I’d appreciate any pointers!

Thanks in advance!

[chenfei-wu]

Hi @user4042025 , thank you for your recognition of our work! Currently, we have only open-sourced the text-to-image foundation model, but the editing model is also on our roadmap and planned for future release.

[user4042025]

Thank you! We’ll be looking forward to the release

[Angericky]

Thank you for sharing your impressive work!
I’ve been also looking forward to exploring the editing capabilities once they’re available.
and I was wondering if you could share any rough timeline or planned release window for the editing model?

It would help us keep an eye out for it. Thanks again for your hard work on this!

[throttlekitty]

I wouldn't have known if it weren't for this. The model card for Qwen Image states the single (and currently released) model can do everything.

[mimimi999]

I believe that Qwen-Image's image editing functions can do the same thing as the Flux kontext model.

If you provide that code in the future, I would also like you to make public the training code for LORA, which uses paired images for style conversion.

FLux kontext Dev's LORA, which performs arbitrary style conversion using paired image learning, is very powerful because it can convert anything and everything into any style. However, there are restrictions on commercial license use, so if similar LORA training could be done with Qwen-Image, it would be possible to migrate existing FLUX users to Qwen-Image.

[matmult]

Is there an image editing API available at least?

[SteveBaumann]

Impressive work. Thank you for the time and effort put into developing a quality model.
It will be exciting to see where the next steps lead to.

[gcalmettes]

Great work on the release! Glad to hear the editing model is also on its way!

[cugzhengzhimin]

I believe that Qwen-Image's image editing functions can do the same thing as the Flux kontext model.

If you provide that code in the future, I would also like you to make public the training code for LORA, which uses paired images for style conversion.

FLux kontext Dev's LORA, which performs arbitrary style conversion using paired image learning, is very powerful because it can convert anything and everything into any style. However, there are restrictions on commercial license use, so if similar LORA training could be done with Qwen-Image, it would be possible to migrate existing FLUX users to Qwen-Image.

I've tried to apply the training method of Kontext to Qwen Image, but the results are far worse than Flux-Dev. Qwen Image seems to struggle with the concatenation of additional tokens, while Flux demonstrates remarkable flexibility, capable of generating well under various condition injection methods. For now, I can only hope for the official editing model from Qwen Image.

[46cv8]

Is there a new model or did someone just figure out a way to do do this using a workaround with the existing model?
huggingface/diffusers#12117

[vladmandic]

Is there a new model or did someone just figure out a way to do do this using a workaround with the existing model? huggingface/diffusers#12117

there's a difference between image-editing model which takes prompt as instructions vs simple img2img. think flux.1-dev can do imgi2mg but flux.1-kontext is actual image-editing model.