Stable diffusion uses a lower-dimensional latent space instead of the actual pixel space to speed up the image generation process. The autoencoder compresses the image into the latent space and ...