Diffusion Model

A generative AI architecture that creates images, video, or audio by learning to reverse a noise-adding process—starting from pure noise and iteratively denoising to produce coherent output. Diffusion models power leading image generators (Stable Diffusion, DALL-E 3, Midjourney) and video generators (Sora). Key variants include latent diffusion (operating in compressed space) and diffusion transformers (DiT).