Course 7, lesson 61 of 100, Ages 10+
Pictures from words
From noise to picture
Like I’m 5
Imagine a fuzzy TV screen slowly clearing until you see a picture of exactly what you asked for. That’s how many image makers work.
The big idea
Many image generators start with a picture of pure random noise, like TV static. Step by step, the model removes a little noise, guided by your words, until a clear picture appears.
It learned how by studying millions of images with captions: during training, it practised turning noisy images back into clean ones. That's why it can paint 'a cat astronaut in watercolour', even though no such photo exists.
Examples
- From static: Noise slowly becomes a picture that matches your words.
- Styles: 'In watercolour' or 'as a pencil sketch' changes the look.
- Odd hands: Generators sometimes draw extra fingers, so check details.
Check your understanding
- How do many image makers start a new picture?
- Options: From random noise; From a photo of you; From a blank page of text.
Answer: From random noise. Diffusion models start with noise and clean it up step by step. - What does an image generator often start from?
- Options: Random noise; A finished photo; A blank wall.
Answer: Random noise. It removes noise step by step, guided by your prompt.
Remember
Image makers turn noise into pictures, guided by your words.
Talk about it
If you could make any picture with AI, what would it be?
Go deeper
This mirrors diffusion models, which learn to remove noise step by step, guided by the text prompt. It’s a good moment to talk about honesty: if AI helped make something, say so.