In this chapter, we first take a look of some unsupervised deep learning models, namely the autoencoders, and explore their use in a few interesting applications. The self-reconstruction engines are made more interesting by taking up some latent representation in a variational configuration, allowing new representations to be sampled from a probability distribution, which are then passed to the decoding stream to generate new patterns. Then, we proceed to another very interesting and important generative model that incorporates adversarial learning, literally allowing rival components in the network to learn from mistakes and up their game, leading to enhanced generative capabilities.