search

kongzhecn / OMG

[ECCV 2024] OMG: Occlusion-friendly Personalized Multi-concept Generation In Diffusion Models
https://kongzhecn.github.io/omg-project/
559 stars 38 forks source link

About prompt to prompt implementation. #12

Closed Adenialzz closed 1 month ago

Adenialzz commented 1 month ago

Hi, thanks for sharing.

I have one question. What did width and height here do? It seems different from the original implementation of prompt-to-prompt.

kongzhecn commented 1 month ago

In gradio demos, like this, OMG supports generating images with different resolutions. Hence, the size of the attention map should vary with the resolution of the generated images.

Adenialzz commented 1 month ago

got it. thanks