AI image glossary
What Is a Diffusion Model?
Diffusion models create images by learning to reverse noise — the technology behind modern AI art. A plain-English explanation of how they work.
The plain-English answer
A diffusion model is the type of neural network behind most modern AI image generators. It learns by destroying millions of training images with progressive noise and practicing how to reverse the damage — so when you give it pure random noise plus a text prompt, it can "denoise" that static step by step into a brand-new image matching your description.
The counterintuitive part is that generation is subtraction: the image isn't painted stroke by stroke, it's revealed by repeatedly removing noise in the direction your prompt points. Twenty to fifty denoising steps take it from static to photograph.
This is why the same prompt yields different images (different starting noise), why a "seed" makes results reproducible (same starting noise), and why prompts guide rather than dictate — the model is steering a controlled accident.
Put it into practice
See a diffusion model work — from prompt to picture.
