This research explores computer generation of artwork using neural networks and proposes a step towards defining creativity in machines. Computer creativity is evaluated using two criteria: novelty and quality. Novelty is the extent to which the artifact produced is different from the training data, and quality is defined as the extent to which it satisfies the definition of the class of artifacts it should belong to. Using the StyleGAN architecture as a starting point, this work is able to generate landscapes with a measurable degree of novelty and quality by learning directions in the latent space. By directly shaping the intermediate latent space, the model generated predictable scenes with chosen labels such as “tree” and “ocean”. The latent vectors of the generated art were projected into three-dimensional space and the trajectories of the transformations were visualized. This intermediate latent space was also used to measure the distance between images. This distance provides a measurement of novelty, and thus creativity because it reveals the extent to which the artifact produced is different from the training data. The artwork produced was analyzed and labelled by pre-trained vision models and the subjective quality of the art was rated by human participants.