extern/easydiffusion
1
0
Fork 1
mirror of https://github.com/easydiffusion/easydiffusion.git synced 2025-06-20 18:08:00 +02:00
Code Issues Packages Projects Releases Wiki Activity

Created VAE - Variational Auto-Encoder (markdown)

JeLuF 2022-10-29 09:50:05 +02:00
parent 64b6e8328b
commit 0149996a8d
1 changed files with 12 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
VAE---Variational-Auto-Encoder.md Normal file

@ -0,0 +1,12 @@
🔥🔥This feature is currently only available on the beta branch, which might be instable and break your installation!🔥🔥
## What does a VAE do?
In short, it improves the generated images. A VAE is trained for certain aspects of the image, and the default VAE bundled in our UI improves the eyes in generated images.
Stable Diffusion UI installs and starts using the `vae-ft-mse-840000-ema-pruned` VAE file that works with all Stable diffusion models. This uses the official one: https://huggingface.co/stabilityai/sd-vae-ft-mse-original
You have two ways to use a custom VAE file:
1. Let's say you're using a model file named `my-model-1.ckpt`. You can now place a VAE file for that model, with the same name (i.e `my-model-1.vae.pt`) inside `models/stable-diffusion`, next to your `my-model-1.ckpt` file. The custom VAE file should have the same name, with `.vae.pt` instead of `.ckpt`. This will result in this VAE file being used (instead of the default) whenever you use `my-model-1` to generate your image.
2. You can place any number of VAE files ending with `.vae.pt` inside `models/stable-diffusion`, and refresh your browser UI. After that, you can set the Default VAE in System Settings (top-right corner of the UI). This VAE will now be used by default, unless a model has custom VAE (set in option 1, described above).