Decoder. plain-English AI glossary

Diffusion

● Core

Iteratively adding noise to images then learning to reverse the process, generating sharp images from random noise.

Think of it like

Rewinding a video of fog rolling in until you see the landscape appear from mist.

Example

Start with random noise, denoise 50 times, each step reveals more detail until a clear image emerges.

How it actually works

Markov chain that gradually adds noise to images until they are pure noise; training objective learns reverse process (denoise). Forward process and reverse process are both Gaussians; posterior can be computed analytically. Conditioning (text, image, class) guides denoising. Sampling uses DDPM, DDIM, or Euler steppers.

For product teams

Most reliable generative model; consistent quality; can be run locally; becoming standard for content generation.

For engineers

UNet backbone predicts noise at each timestep; train with MSE loss on predicted noise; sample with iterative denoising.

Image Generation — Used in.
GAN — Alternative to.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open Decoder Add to Chrome