Diffusion Model

A generative model that learns to reverse a gradual noising process, turning random noise into an image step by step.

Generative AI Advanced 1 min read

Definition

Training a diffusion model means taking real images and progressively adding noise until nothing is left, then training a network to undo one step of that process. Generation runs the learned reversal from pure noise, denoising repeatedly until an image appears.

Conditioning the denoiser on a text embedding is what makes it follow a prompt. Compared with GANs, diffusion models train far more stably and cover the data distribution better, at the cost of needing many steps to sample — which is why most of the engineering effort goes into reducing the step count.

How it relates to other terms

Contrast with Generative Adversarial Network Two networks trained against each other — one generating fakes, one detecting them — until the fakes are convi... Prerequisite Embedding A list of numbers representing a piece of data, arranged so that similar things end up close together.
Referenced by
Generative Adversarial Network Contrast with

Related terms

Prompt Engineering

Writing the input to a model deliberately — with context, examples and constraints — to get a more reliable output.

Beginner Generative AI

Temperature

A setting that controls how much randomness goes into choosing each token — low is predictable, high is varied.

Beginner Generative AI

Was this definition helpful?

Definitions are reviewed by our editorial team. Spotted a problem? Tell us.

Keep reading

Browse the glossary Suggest a correction