A DL project that implements Neural Style Transfer on images using a pretrained model, a VGG19-based architecture, and a Feed-Forward architecture after Johnson et al. (2016). The different models utilize custom TensorFlow training loops, cost functions and layers (InstanceNormalization). Various tendencies of such architectures are discussed. The results are satisfying, yet further improvements are always possible.