Generative Adversarial Networks have been shown to produce realistic and high quality images by leveraging adversarial training between a generator network and a classifier network. We implement and analyze two different image to image translation GANs, Pix2Pix and CycleGAN, on our datasets with the goal of producing highly interpretable black and white sketches. We analyze the performance of these GANs on different datasets and were able to assess some of the strengths and weaknesses of Pix2Pix and CycleGAN. For Pix2Pix, an improved generator architecture, data augmentation effect and training guidelines were also suggested.