Can you show an example of where the current code (without the changes in this PR) fails?

from diffusers import PixArtAlphaPipeline, Transformer2DModel
import torch

pipe = PixArtAlphaPipeline.from_pretrained("PixArt-alpha/PixArt-XL-2-1024-MS", 
                                           torch_dtype=torch.float16).to('cuda')
prompts = ["a man", "cat riding a bike in the forest, watercolor style"]   
images = pipe(prompt=prompts, num_images_per_prompt=2).images

images[1] and images[2] will be generated with wrong prompt_attention_mask.

Thanks! Could we add a fast test for this as well?

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

hi @nikitabalabin
thanks for the PR! let us know if you're still interested in finishing it:)

can we run make style too?

@hlky was the close expected?

Rebase to fix conflict went wrong 😅