Phillip Isola introduces generative modeling in this lecture from MIT's 6.7960 Deep Learning course. He lays out the core mathematical framing of density and energy-based models before moving to the practical question of how to sample from a learned distribution. The lecture then surveys the major model families that solve this problem in different ways: generative adversarial networks, which pit a generator against a discriminator, autoregressive models, which factor a distribution into a sequence of conditional predictions, and diffusion models, which learn to reverse a noising process. Isola connects the underlying theory to how each architecture is trained and what tradeoffs it makes between sample quality, tractable likelihoods, and computational cost. Running 81 minutes, it functions as the conceptual foundation for the course's later, more specialized treatments of these model classes.