1. The physics analogy explains how diffusion models work using principles from thermodynamics and statistical mechanics. 2. These models treat pixels like atoms, which can randomly move to form "TV static". 3. By modeling this random motion in reverse time, we learn how to generate novel images with the same characteristics as our training data. 4. This process of generating new images from a given text prompt is used by text-to-image models like DALL-E 2 and Imagen.