Diffusion model
A diffusion model is a generative model that learns to turn random noise into data, such as an image, by removing a little noise at a time over many steps.
Also called: diffusion, denoising diffusion, latent diffusion
By Divya Sharma · Checked Nov 15, 2026 · How we review
Training starts with real images and adds noise to them step by step until nothing is left but static. The model learns to undo one step of that noise. To generate, you start from pure noise and run the learned denoising step again and again until a clean image appears. The 2020 DDPM paper by Ho, Jain and Abbeel showed this approach could produce high-quality images.
Why it matters
Most image and video generators you meet today, from text-to-image tools to text-to-video models, are built on diffusion. Open examples include FLUX.2 dev, and libraries such as Diffusers and ComfyUI run them locally. More models are listed under image models.
Latent diffusion
Running diffusion directly on pixels is slow and costly. Latent diffusion, introduced by Rombach and colleagues, first compresses the image with an autoencoder and runs the denoising in that smaller space. Cross-attention layers let a text prompt steer each step. This is the design behind Stable Diffusion and many models since.
Limits
Generation needs many sequential steps, so it is slower than producing an image in one pass. Steering is indirect: the prompt, the guidance scale and add-ons like ControlNet all nudge the denoising rather than dictate the result. A diffusion language model applies the same idea to text.
Related terms
- Text-to-image
- Text-to-image is the task of generating a new picture from a written description, called a prompt. Most current text-to-image models are diffusion models conditioned on the text.
- Text-to-video
- Text-to-video is the task of generating a short video clip from a written prompt. Models must produce frames that each look right and also move consistently from one frame to the next.
- Guidance scale (CFG)
- The guidance scale, or CFG scale, is a setting in diffusion image generators that controls how strongly the output follows the prompt. Higher values follow the prompt more closely but reduce variety.
- ControlNet
- ControlNet is a neural network add-on that lets a text-to-image diffusion model follow a structural guide, such as an edge map, depth map or human pose, in addition to the text prompt.
- Diffusion language model
- A diffusion language model (dLLM) generates text by starting from a fully masked or noisy sequence and refining many token positions in parallel over several steps, instead of writing one token at a time.
- Autoregressive model
- An autoregressive model generates a sequence one piece at a time, predicting each next token from all the tokens before it and feeding that output back in. Most large language models work this way.