Hi,
Thank you for sharing this great work!
I tried running the provided Flux example exactly as in the repository, but most of the attention maps I obtained were completely black.
Only a few timesteps and layers produced meaningful attention maps. Is this the expected behavior?
Also, I used the same example prompts shown in the README, but my results look quite different from the examples provided there.
Could you please let me know what might be causing this difference?
Thanks a lot for your time and help!