What is Latent Space?

How it works

Latent space is the compressed representation that AI models use internally. Instead of working directly with pixels (which would be computationally prohibitive), diffusion models compress images into a much smaller latent representation, perform the diffusion process in this compressed space, then decode the result back into pixels. This is why they are called 'latent diffusion models'. The latent space captures semantic meaning: nearby points in latent space represent visually similar content. This means you can interpolate between two points in latent space to create smooth transitions, which is how some AI video tools generate motion.