How It Works
Diffusion Model
The technique behind most modern image generators: starting from random noise and gradually “denoising” it into a coherent image, reversing a process trained by progressively adding noise to real images.
Origin
Theoretical foundation from Jascha Sohl-Dickstein and colleagues (2015); the modern practical breakthrough is Jonathan Ho, Ajay Jain, and Pieter Abbeel’s 2020 “Denoising Diffusion Probabilistic Models” paper.