Diffusion model
A generator that starts from random noise and removes it step by step until an image, video or sound matches the prompt.
To train a diffusion model, you take millions of real images, add noise to each a little at a time until it's pure static, and teach a model to undo one step of that. To generate, you run it backwards: start with fresh static, and let the model remove noise in many small steps, guided by your prompt, until a clear picture is left.
Many well-known image generators, such as Stable Diffusion and Google's Imagen, work this way, and the same idea is used for video and audio. Google DeepMind has also shown an experimental text model built on it, Gemini Diffusion, which writes by refining a whole passage at once rather than word by word.
For example
Ask for "a red bicycle against a blue wall" twice and you get 2 different bicycles: each run started from different random noise.
Put it to use
Where you'll meet diffusion model in ChatGPT, Gemini and Claude, how to use it there, and the mix-up most people make.
Daily Bytes Plus is in the Android app.