Generative models learn an underlying data distribution, which can be sampled to produce new instances similar to the training data. Examples of this include creating pictures of made-up buildings or portraits of people that don’t exist. To facilitate the research in this field, researchers at the Chinese University of Hong Kong have created GenForce, a PyTorch library that includes fast, distributed, and highly reproducible training code and a model zoo with many pre-trained variants of StyleGan, PGGAN, and other models.