A generative model that learns to create data by reversing a noising process, starting from random noise and iteratively denoising to produce realistic samples.