# What is VQ-VAE? Definition and meaning

A variational autoencoder that quantizes its latent space into a learned codebook of discrete tokens.

## Definition

**VQ-VAE**: A variational autoencoder that quantizes its latent space into a learned codebook of discrete tokens.

In short: Autoencoder with a discrete latent codebook.

## When it is used

Tokenizing images and audio for autoregressive and diffusion models.

## Learn more

The full guide to **[VQ-VAE](/tools/vq-vae)** explains how it works, when to use it and how it compares with the alternatives, with examples. The [glossary](/glossary) lists every term on the site from A to Z.

Previous term: [vLLM](/glossary/vllm) · Next term: [Weaviate](/glossary/weaviate)
