stable-diffusion-webui icon indicating copy to clipboard operation
stable-diffusion-webui copied to clipboard

Published 20 hours ago • verified publisher icon AUTOMATIC1111

Cycle Diffusion. img to img composition in SD

Open jvvilar opened this issue 3 years ago • 3 comments

https://github.com/ChenWu98/cycle-diffusion

A method of infering the random seed from a real image, causing SD to generate an identical one, which can then be composed differently by changing specific parts of the prompt to a target text.

jvvilar avatar Oct 16 '22 03:10 jvvilar

seems intresting

TingTingin avatar Oct 16 '22 07:10 TingTingin

Goods news: I got cycleDiffusion running in automatic

https://github.com/nagolinc/auto_cycleDiffusion/tree/main

Bad news: It uses a ton of VRAM (can't run in 16gb free colab)

nagolinc avatar Oct 22 '22 15:10 nagolinc

Self Attention Guidance is a big improvement for small details in an image without affecting overall composition.

https://huggingface.co/spaces/susunghong/Self-Attention-Guidance

Since they like to set their CFG values weirdly low in the samples, here's my own example with it set to 10:

https://imgur.com/a/zypdU5D (The bottom image has the new feature enabled)

It just looks a lot cleaner.

lendrick avatar Feb 19 '23 18:02 lendrick