Generative Adversarial Networks (GANs) synthesize content with features similar to the original training set, and can be conditioned on specific inputs. For example, GANs could be applied to generate a new set of sneaker designs from a set of hundreds of shoe images. To make it easier to use GANs, NVIDIA has released Imaginaire, a Pytorch library containing optimized implementations of several GAN-based image and video synthesis methods (including both supervised and unsupervised approaches).