search

google / style-aligned

Official code for "Style Aligned Image Generation via Shared Attention"
Apache License 2.0
1.19k stars 86 forks source link

support for stable diffusion #1

Closed Tord-Zhang closed 8 months ago

Tord-Zhang commented 8 months ago

Hi, wonderful work!

The code provided is for SDXL, would it work for SD? since the UNet in SD and SDXL are different.

Another question is when using with controlnet, the paper only show examples where prompts are different and control is the same. What about the same prompt with different controls? Would it works?

matrix4767 commented 8 months ago

Asking for this as well, for SD 1.5.

amirhertz commented 8 months ago

Added notebook for SD 1.* It works the same, but the quality is not as good as SDXL.

Using different controls with the same prompt should work, you only need to fix the seed for the reference image: fix the provided latent of the first image in the batch.

Tord-Zhang commented 8 months ago

Added notebook for SD 1.* It works the same, but the quality is not as good as SDXL.

Using different controls with the same prompt should work, you only need to fix the seed for the reference image: fix the provided latent of the first image in the batch.

the style aligned images are much worse than that of SDXL, any insight why?