A diffusion model is used to remove noise from images or to generate realistic images incduing deep fakes. The diffusion model is trained by adding more and more noise to an image and then presenting the noisy and original image as input-output pairs to a deep neural network.
Defined on page 280
Used on Chap. 12: pages 280, 281
Training a diffusion model.
Using a diffusion model to generate images from prompts.