A generative AI architecture that creates outputs by iteratively denoising random noise, commonly used for image generation. Diffusion models power DALL-E, Midjourney, Stable Diffusion; they raise distinctive copyright and provenance questions for images.