Diffusion models, visualized
Diffusion models generate images by learning to reverse a process of adding noise. In the forward direction, a clean image is gradually corrupted into pure static. The model is trained to undo one step of that corruption at a time.
To create a brand-new image, it starts from random noise and denoises step by step until a coherent picture emerges. This is the engine behind Stable Diffusion, DALL·E, and Midjourney.
Scrub the noise level (forward ⇄ reverse)
Noised 40%
in between
Each real step adds/removes just a little noise; there are hundreds of them.
Free · runs entirely in your browser · nothing to install
How to use it
- Scrub the slider forward to watch a clear image dissolve into noise.
- Scrub back to watch the reverse (denoising) process reconstruct it.
- Notice that generation is just this reverse process starting from pure noise.
What you'll take away
- The forward (noising) vs reverse (denoising) processes.
- Why generating an image means removing noise step by step.
- How the same idea powers today's top image generators.
Want to actually build this?
This demo is one moment inside a full Math to Machine lesson — predict, build, and explain the concept, with an AI tutor that gives hints, not answers. The first five lessons are free.
FAQ
- How do diffusion models generate images?
- They start from random noise and repeatedly remove a little noise, guided by what they learned during training, until a clean image appears. Training teaches them to reverse a gradual noising process.
- What models use diffusion?
- Most modern image generators, including Stable Diffusion, DALL·E, and Midjourney, are based on diffusion. The same principle is being extended to video and audio generation.