--zero_cond_t # This is a special parameter introduced by Qwen-Image-Edit-2511. Please enable it for this model.

[lisi31415926]

https://github.com/modelscope/DiffSynth-Studio/blob/main/examples/qwen_image/model_training/lora/Qwen-Image-Edit-2511.sh --zero_cond_t # This is a special parameter introduced by Qwen-Image-Edit-2511. Please enable it for this model.

[willhsmit]

I think, if I understand it correctly, that the way AI-toolkit works, it is up to the diffusers library to identify the 2511 model as requiring zero_cond_t based on the config.json model.

I don't think AI-toolkit needs to do anything to explicitly enable the parameter in code, but it does need a version of diffusers library that is newer than the 2511 fixes in https://github.com/huggingface/diffusers/commit/b8a4cbac14d32afa6c6e6c5b9cd17f9715214220 .

[lisi31415926]

I think, if I understand it correctly, that the way AI-toolkit works, it is up to the diffusers library to identify the 2511 model as requiring zero_cond_t based on the config.json model.

I don't think AI-toolkit needs to do anything to explicitly enable the parameter in code, but it does need a version of diffusers library that is newer than the 2511 fixes in huggingface/diffusers@b8a4cba .

https://github.com/ostris/ai-toolkit/blob/main/requirements.txt git+https://github.com/huggingface/diffusers@6bf668c4d217ebc96065e673d8a257fd79950d34

it means right?

[willhsmit]

Yes, I replaced that @6bf...d24 string in requirements.txt with the latest commit (f6b6a7181eb44f0120b29cd897c129275f366c2a) locally.

I haven't finished baking my lora yet but the sample gens - both before and after training - look much better with the newer commit than the old. And, updating the diffusers line didn't appear to break any compatibility or require other packages to be changed in requirements.txt.

[lisi31415926]

Yes, I replaced that @6Bf...d24 string in requirements.txt with the latest commit (f6b6a7181eb44f0120b29cd897c129275f366c2a) locally.

I haven't finished baking my lora yet but the sample gens - both before and after training - look much better with the newer commit than the old. And, updating the diffusers line didn't appear to break any compatibility or require other packages to be changed in requirements.txt.

Thank you for your reply. Also, how did you process the controlled images? Were they black images?

[lisi31415926]

Yes, I replaced that @6Bf...d24 string in requirements.txt with the latest commit (f6b6a7181eb44f0120b29cd897c129275f366c2a) locally.

I haven't finished baking my lora yet but the sample gens - both before and after training - look much better with the newer commit than the old. And, updating the diffusers line didn't appear to break any compatibility or require other packages to be changed in requirements.txt.

https://github.com/huggingface/diffusers/pull/12839
Maybe help to understand~

[lisi31415926]

https://huggingface.co/Qwen/Qwen-Image-Edit-2511 guide: Install the latest version of diffusers
pip install git+https://github.com/huggingface/diffusers