AI Glossary

Latent diffusion

Also called LDM

What does “Latent diffusion” mean?

Latent diffusion: Running the diffusion process in a compressed representation (the "latent space") rather than on full-size pixels. An autoencoder shrinks images into compact latents, diffusion happens there, and a decoder turns the result back into an image. This makes generation far cheaper.

Latent diffusion visual guide: Running the diffusion process in a compressed representation (the "latent space") rather than on full-size pixels. An autoencoder shrinks images into compact latents, diffusion happens there, and a decoder turns the result back into an image. This makes generation far cheaper.
Visual guide Open full size (new tab)

What is an example of “Latent diffusion”?

Working in latents is a key reason Stable Diffusion could run on consumer graphics cards rather than only in data centers.

Keep learning

See also

Diffusion modelTermLatent spaceTermVariational autoencoderTermText-to-imageTerm

Same topic

Lip syncTermInpaintingTermLoRA (image models)TermImage-to-videoTermMusic generationTerm