Generative Adverserial Networks (GANs)  have established themselves in the last couple of years as a staple in deep generative models. They provide state of the art results and inception scores  on many of the well studied and experimented on datasets such as Cifar100 in and LSUN-Bedrooms in . In Context of this Bachelor thesis I implement the Wasserstein GAN model  on a new dataset that I extract from a hand painted film. I test the robustness of my network to gain further insight into the necessary requirements for not only the stability of the model but also good quality generated images. Through trial and error I establish best practices relating to the choice of hyperparameters and the degree of freedom the model allows. I also adopt a hierarchical approach where I create different subsets of our training data and compare the generated images from each subset. This allows me to draw conclusions about which training data delivers the best performance, assumptions about the behaviour of the model and ideas for future research.