← Glossary · Architecture

diffusion model

Technique

Fact-checked May 20, 2026

Also called: diffusion models, denoising diffusion probabilistic model, DDPM

Diffusion models are a type of generative AI model that learn to create new data, like images or audio, by progressively removing noise from an initial random input.

Imagine taking a clear image and slowly adding static or noise until it's just random pixels. A diffusion model essentially learns to reverse this process. It starts with pure noise and, through a series of steps, gradually 'denoises' it until a recognizable image or other form of data emerges.

This technique has become incredibly powerful for generating realistic images, like those you see from tools such as DALL-E 2 or Midjourney. It works by training a neural network to predict and remove noise at each step, guided by text prompts or other inputs. The iterative nature of this denoising process allows for high-quality and diverse outputs that closely match the desired description.

Beyond image creation, diffusion models are also being explored for tasks like audio generation, video synthesis, and even drug discovery, showing their versatility in learning complex data distributions.

Learn AI in 5 minutes a day.

Daily Deck explains terms like diffusion model as part of a free seven-card daily brief. No jargon. No fluff.

Start free