A diffusion model generates an image by starting from pure noise and gradually denoising it over many small steps until a coherent image emerges. The model learns this denoising process by reversing a forward process that added noise to real training images. Almost every major AI image model in 2026 — Stable Diffusion, Midjourney, Flux, DALL-E, Imagen, Nano Banana, Seedream — is a diffusion model or a close descendant. The main exception is autoregressive models like the older Parti family.