Latent Diffusion

Latent Diffusion is a generative modeling approach that runs a diffusion process in a compressed latent space rather than directly in the high-dimensional data space. Originally developed for image synthesis in models like Stable Diffusion, the approach can be adapted to text by first encoding discrete tokens into continuous latent vectors, then training a diffusion model to learn their distribution — making generation both more tractable and more amenable to continuous-space techniques.