How Diffusion Models Generate Images
Part of the Prompt Engineering course
Until now every prompt you wrote produced text. This lesson opens Part 3, where the same prompt produces a picture. The medium changes but the mindset from Part 1 does not: you are still steering a model with words, and vague words still get you vague results. Before you can prompt an image model well, it helps to know what it is actually doing when it turns your sentence into a picture.
A common guess is that an image model searches a huge library and stitches together pieces of existing pictures. That is not what happens. There is no collage and no lookup. The model builds every image from scratch, and the raw material it starts from is pure random noise.
Want to learn Prompt Engineering How Diffusion Models Generate Images in a real environment?
Click below, get a fresh Linux box with Prompt Engineering ready to go, and work through How Diffusion Models Generate Images hands-on with an AI tutor that knows your code.
Start this lessonHow Diffusion Models Generate ImagesWhat you'll learn
- Diffusion as iterative denoising
- Text conditioning of the image
- Seeds and reproducibility
- Sampling steps and guidance scale