Generative model
A generative model learns p(x), or p(x, y), rather than only a decision boundary p(y | x) as a discriminative model does. Sampling from it creates novel data; families include autoregressive models, VAEs, GANs, normalizing flows and diffusion models.
Example
Trained on photos of faces, a generative model can sample a face that appears nowhere in the dataset, while a discriminative classifier trained on the same photos can only say "face" or "not face".