potential Qwen Image batching issue

[dxqb]

Hi @kohya-ss
I haven't confirmed - still downloading the models and learning how to set everything up, because I want to try Qwen Image Edit training - but it looks like your code is affected by this bug:
huggingface/diffusers#12294

It can lead to incorrect conditioning if batch size > 1
A workaround is posted in the diffusers issue. Or, since you've copied the pipeline into your repo, it could also be fixed there directly.

[kohya-ss]

Thank you for letting me know.
This bug does indeed affect this repository. I didn't notice that encoder_hidden_states_mask wasn't being used when I copied the code.
I'll fix it as soon as possible.