Diffusion-based probabilistic models fall into the Artificial Intelligence category and are particularly used in the field of modern image and speech generation. It is an innovative method that allows computers to create images, texts, or even music that appear deceptively real.
The principle is comparable to the creation of a photograph from a multitude of pixel disturbances. Initially, the model takes „noise“, i.e., a chaotic, random image, and then brings order to it step by step. In each step, the model estimates how the image could look more realistic, until finally a clearly recognisable image emerges.
A practical example: If someone wants to generate a photorealistic image of a green cat in a tree, the diffusion-based probability model starts with a random pattern. After many small improvements, the final result actually shows a green cat in a tree – even though no such image existed before.
The advantage of this technique is its ability to generate new and astonishingly realistic content. This opens up exciting possibilities in advertising, product development, or design for bringing creative ideas to life.
Deutsch 
English (UK)