AI Literacy / Glossary

Diffusion model

A diffusion model is a type of generative AI that makes images (and sometimes video or audio) by starting from random noise and removing it step by step, guided by a text prompt, until a picture appears.

During training, the model sees many images with more and more noise added, and learns how to undo that noise a little at a time. It also learns which image features go with which words.

To create a new picture, it starts with pure static and repeatedly predicts a slightly less noisy version that fits the prompt. After many small steps, a detailed image emerges.

Because the model rebuilds images from learned patterns, it can produce typical mistakes, such as odd hands, garbled text or impossible reflections, though these are getting rarer. That is why labels, provenance information and context matter more than spotting glitches.