Architecture
Diffusion Model
A generative model that creates images (or other data) by learning to reverse a gradual noising process, producing high-quality outputs from random noise.
Explained at five levels
Level 1
The AI that makes pictures — it starts with TV static and slowly turns it into a beautiful image, like magic.
Level 2
The type of AI behind image generators like DALL-E and Midjourney. It learns by adding noise to images, then learns to remove it to create new ones.
Level 3
A generative model that creates images (or other data) by learning to reverse a gradual noising process, producing high-quality outputs from random noise.
Level 4
A generative architecture that learns a denoising process over a forward diffusion (noising) trajectory, enabling high-fidelity image, video, and audio synthesis via iterative refinement from Gaussian noise.
Level 5
A latent-variable model defining a forward Markov chain that gradually adds Gaussian noise and a learned reverse process that denoises — optimized via a variational bound equivalent to denoising score matching, with classifier-free guidance steering generation.
Definitions are educational summaries. Terminology can vary by source and context.
Sources
- NIST Trustworthy and Responsible AI Resource Center: Glossary — National Institute of Standards and Technology. Accessed 2026-07-20.