Glossary
Diffusion Model
The class of generative AI that creates images by repeatedly de-noising random static.
A diffusion model is trained to reverse a noising process: start with pure random noise and, step by step, predict what to subtract to reveal the image described by your prompt. Stable Diffusion, Flux, DALL·E 3 and Midjourney all use this approach.
Diffusion replaced GANs as the dominant image-generation architecture around 2022 because it produces more diverse, higher-fidelity results from text.