new explain_image notebook #16
Conversation
also: - pass latent through run_on_prompt - add more scaling options to show_cross_attention, etc. - puppy image generated using StableDiffusion 2.1 in DreamStudio - need latest pytorch (in requirements.txt) for RTX4090
|
Hi, I adapted your (very useful!) code into a new notebook to make it easier to visualize the attention maps of an arbitrary image and a specified text prompt. Maybe this can help others too. |
|
I also verified that the |
|
Thanks @claforte for the neat visualizations! There are some things we noticed that we want to consider before merging the pull request. First, the code does not seem to work with the new diffusers version (0.12.1) and we want to make sure our code remains updated with its latest versions. We also noticed that you updated the torch version, and we want to make sure that there are no compatibility issues with the updated torch version and the other packages. Finally, our code does not official support Stable Diffusion 2.1 and since the image you provided here is outputted with SD2, we prefer to wait until we officially add support for SD2 and ensure that our code works with the new model. |
also:
Screenshot:
