In plain English
A diffusion model is a type of generative AI that creates images, and other data, by starting from random noise and gradually removing it, step by step, until a coherent result emerges. A text or image prompt guides the process so the final picture matches what was asked for. Diffusion is the technique behind many popular image generators.
A worked example
Tools such as Stable Diffusion and DALL-E turn a prompt like 'a watercolour fox in a snowy forest' into a new image by denoising a random starting canvas.
Common confusion
A diffusion model does not search for or paste together existing pictures. It generates a new image from noise, shaped by the patterns it learned during training.