Glossary · technique
Diffusion Model
Also known as: diffusion AI, latent diffusion
Definition
A diffusion model is a type of generative AI that creates images or video by progressively denoising random noise, guided at each step by a learned prediction of what the final image should look like.
In more detail
Diffusion models power most modern image and video generators (Flux, Stable Diffusion, Imagen, Veo). They work by reversing a learned "noising" process — adding noise to training images, then training the model to reverse it. Modern variants like flow-matching and rectified flow accelerate this from ~50 steps down to 1–4 for fast variants like Flux Schnell.
Try it on Mobbi
FAQ
Are LLMs diffusion models?+
No — LLMs like GPT-4 and Claude are autoregressive transformers. Diffusion models are a different architecture used for images, video, and increasingly audio.