Manga colorization is a challenging and time-consuming task for creators. This thesis aims to establish a building block for coloring sketches- es/manga when given a sketch as input. Approaches for utilizing a data source are explored. The thesis explores pix2pix GAN(Conditional Generative Adversarial Network) and its application to the use case. Also, other experiments that include altering the discriminator architecture and altering the generator loss function are explored. Finally, all the models are hosted live on hugging face to visualize comparisons across experiments.