This repository provides pure implementations of generative adversarial network architectures as described in their original papers, including DCGAN, LSGAN, WGAN, WGAN-GP, EBGAN, BEGAN, DRAGAN, and CoulombGAN. All models are implemented in TensorFlow and trained on the CelebA and LSUN bedroom image datasets at 64x64 resolution. The project aims to provide unbiased comparisons by not cherry-picking results, with training powered by TensorFlow queue runners and metrics logged to TensorBoard.