The task of a generative model is to learn a distribution, and then sample it.