Likelihood-based Generative Models are a class of generative model that model the distribution of the data $p\left(y\right)$ directly with a likelihood function. The most popular subclass of likelihood-based generative models is variational autoencoders.