Diffusion

Diffusion starts with noise and cleans it step by step into an image or video.

What it does

Diffusion starts with noise and cleans it step by step into an image or video.

How it works

• Start with random dots.
• Use a prompt or goal as a guide.
• Clean the dots over many steps.

Uses and applications

• An art tool can make images from text.
• A video tool can make short draft clips.
• A designer can quickly test visual ideas.

In short

Diffusion is the core idea behind many modern image and video generators.

Related methods

• Stable Diffusion: Runs diffusion in a compressed latent space, guided by text.
• GANs: Generates images in one pass, using a generator and a judge.
• VAE (Autoencoders): Another generative model built around a compact code.

Back to Image and content generation · All topics

The interactive demonstration and the Python walkthrough for this topic need JavaScript.