Latent Diffusion
Definition
Latent diffusion performs the diffusion process in a compressed latent representation space instead of directly on pixels.
An autoencoder compresses images; the diffusion model operates on these latent representations and then decodes them back. This can reduce computational effort.
Simply explained
How to first encode an image compactly, generate there, and then unpack it again.
Why it matters in practice
Relevant for modern image generation and understanding many diffusion models.
Example from business
An image model generates variants in a latent space and then decodes the result.
Limits and pitfalls
- Interpreting latent space as directly understandable image features; equating model families indiscriminately.
Origin & history
Rombach et al. published Latent Diffusion Models in 2021/2022 as an approach for high-resolution image synthesis with reduced computational effort.
Use in mid-sized companies
- Possible application in SMEs: technically evaluating generative image systems and classifying resource requirements.
Frequently asked questions
Is latent diffusion the same as any diffusion approach?
No. Why latent? To compress representation and computational effort.
Sources
- High-Resolution Image Synthesis with Latent Diffusion Models · Rombach et al. (2022), High-Resolution Image Synthesis with Latent Diffusion Models
Sources checked on 10.10.2026.
Editorially reviewed on: