OPENGIT/stable-diffusion-webui
1
0
Fork 0
mirror of https://github.com/AUTOMATIC1111/stable-diffusion-webui.git synced 2025-04-11 23:39:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

add MultiDiffusion with Tiled VAE

ClashSAN 2023-03-08 06:09:01 +00:00
parent 21b03a64ef
commit a707a262b3
1 changed files with 20 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
Extensions.md

@ -18,6 +18,26 @@ As extensions allow the user to install and run arbitrary code, this can be used
# Extensions
## MultiDiffusion with Tiled VAE
https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111
### MultiDiffusion
- txt2img panorama generation, as mentioned in MultiDiffusion.
- It can cooperate with ControlNet to produce wide images with control.
Panorama Example:
Before: [click for the raw image](https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111/blob/docs/imgs/ancient_city_origin.jpeg)
After: [click for the raw image](https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111/blob/docs/imgs/ancient_city.jpeg)
ControlNet Canny Output: https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111/raw/docs/imgs/yourname.jpeg?raw=true
### Tiled Vae
The `vae_optimize.py` script is a wild hack that splits the image into tiles, encodes each tile separately, and merges the result back together. This process allows the VAE to work with giant images on limited VRAM (~10 GB for 8K images!).
Remove --lowvram and --medvram to enjoy!
## VRAM Estimator
https://github.com/space-nuko/a1111-stable-diffusion-webui-vram-estimator