Definition
Plain language
An AI that creates an image by starting from noise and cleaning it up step by step.
As stated in the literature
A generative model that learns to reverse a gradual noising process, producing images or other data through iterative denoising; distinct from autoregressive token generation and central to image-generation stacks like Show-o2.
Also called: diffusion models, diffusion
Why it matters: It's the approach behind much of today's AI image generation, producing detailed pictures from nothing but noise and a prompt.
For example, to make a picture of a cat, the system starts with random static and removes the noise bit by bit until a cat appears.
Heard on the show
“… deploy a model — let's say one of those new multimodal ones that interleaves autoregressive text and diffusion image steps in a single forward pass — and you want to run it on a MacBook. …”Episode 027 — When AI Agents Build the Serving Stack: A Bet on Bespoke Infrastructure