Diffusion Model
In plain terms
A generative model that creates data by reversing added noise.
Going deeper
Diffusion models learn to gradually denoise random noise into coherent data, producing high-quality images and more. They generate by iterative refinement.
Under the hood
Trained to reverse a noising process; behind modern image generators, slower but high-fidelity.
Examples
Stable Diffusion, DALL-E, image generation, audio/video synthesis.
Think of it like...
Like sculpting a statue by progressively removing noise from a block of static.