Generative models that create data by learning to reverse a gradual noise addition process (denoising). They power modern tools like Stable Diffusion.
Inspired by non-equilibrium thermodynamics.
The current state-of-the-art for image, video, and audio generation.