This repository provides a Tensorflow implementation of Wasserstein GAN (WGAN), an alternative to standard GANs that replaces Jensen-Shannon divergence with Wasserstein (earth mover) distance for measuring distribution similarity. The approach addresses training instability issues in traditional GANs by providing a continuous, almost everywhere differentiable metric even when distributions have non-overlapping support. The critic network outputs scores rather than probabilities, enabling reliable gradient signals for the generator.