a visual course

From Noise
to Images

How AI turns random static into a picture — in about half an hour.

A free, visual course on how diffusion models like Stable Diffusion, DALL·E and Midjourney work — from adding noise to an image, to learning to reverse it, to generating pictures from pure random static.

Start from pure noise →

Part 1 — The big idea

1

Destroy, then rebuild

Add noise to a photo until it's pure static — then learn to run it backwards.

2

What noise really is

Gaussian noise, sampling, and why 'random' has a precise shape.

Part 2 — Adding noise on a schedule

3

The forward process

A fixed schedule adds noise over many steps — and you can jump to any step at once.

Part 3 — Learning to reverse it

4

Learning to denoise

A network looks at a noisy image and predicts the noise that was added.

5

Generating from static

Start from pure noise, predict-and-remove a little at a time, and an image appears.

Part 4 — Steering & scaling

6

Steering with words

How a text prompt and guidance push the result toward what you asked for.