mode collapse

[betterze]

Dear Stable Cascade team，

Thank you for sharing this great work, I really like it.

I am playing with the code through huggingface. I find the results have a very strong mode collapse, although it is a regular diffusion (not a few-step model). Could you tell me why this happens?

The following results are with text prompts 'a photo of a dog', 'a photo of a cat', 'a photo of a man'. They all have a black background, a frontal face, the cats have the same color, and the dogs have the same color.

Thank you for your help.

Best Wishes,

Zongze