Diffusion Models
Tools like Stable Diffusion and Midjourney make a picture from a few words. They do it by starting with pure noise and removing it step by step. This lesson shows how a model learns to do that, and how your prompt guides it.
New words
- Noise
- Random dots with no meaning, like the snow on a bad TV signal.
- Denoising
- Removing noise from a picture to get a clearer one.
- Diffusion step
- One small round of noise removal. A picture needs many rounds.
- Guidance
- A push that makes the model follow your prompt more closely.
- Latent
- A small, compressed form of a picture that the model works on.
- Prompt
- The text you write to tell the model what to make.
What you will learn
- How can a model make a picture from words?
- Training: add noise, then learn to undo it
- Making a picture: start from noise and clean it
- How your prompt steers every step
- Try it: clean the noise yourself
- Latent diffusion: work on a small picture
- Trade-offs
- Where it is used, and where to be careful
- Interview tip
- Key takeaways
10 quiz questions at the end.
The full lesson, with interactive demos and the quiz, opens after you sign in with Google. It is free.