1. Basic concepts
VAE is the abbreviation of Variational Autoencoder, and the Chinese name is Variational Autoencoder, which is a generative model based on deep learning. The basic idea of VAE is to encode the input data as a probability distribution in the latent space, and map the random vectors in the latent space back into the original data space through the decoder. The training process of VAE consists of two stages: the training of encoder and decoder. During the training of the encoder, the VAE learns how to map the input data to a probability distribution in the latent space by minimizing the reconstruction error. During the training of the decoder, the VAE learns how to generate raw data from random vectors in the latent space by minimizing the KL divergence. The advantage of VAE is that high-quality samples can be generated, and latent space interpolation and manipulation can be performed, thereby achieving control over the generated data.
2. Function
Generally speaking, it is equivalent to a filter to make the color more vivid and the picture more realistic. Load
VAE without loading VAE
3. Open the VAE Settings-User interface-Quicksettings list in the WebUI
-add ",sd_vae" after sd_model_checkpoint-click "Apply settings" and "Reload UI" to
restart and open the WebUI to see the column "SD VAE", drop down and select the corresponding VAE model can be
4. Storage location of VAE files
Take D disk as an example
D:\stable-diffusion-webui\models\VAE
5. Download VAE files
and open https://civitai.com/
(some models come with VAE)
at https: //civitai.com/ There is no VAE column, you can click to download the corresponding model in the SD-WebUI Launcher-Model Management-Variational Autoencoder (VAE) model. After downloading, the model will be automatically saved to the corresponding
location