In this master’s thesis, the use of an architecture spinning off pix2pix GAN is being investigated for image-to-image translation, transforming segmentation maps into real images. Thesis is split up into several sections: introduction, two sections covering theoretical background, and analysis and conclusion. The first section is the theoretical background, which provides a background on how neural networks and generative adversarial networks work, necessary definitions, and explanations of all the essential components. The second section is the experiments and analysis section, where the dataset is presented, all the experiment parameters and runs, logs, evaluation, and comparison of the novel approach of using CLIP as a loss for image-to-image translation with previous methods on a complex dataset. In conclusion, all the experiments and their results are summarized. In the literature list is all the used literature.