Generative adversarial network (GAN) technology has dramatically advanced over the past 9 years since it was first published. GANs are a type of machine learning infrastructure created from two competing neural networks that can be trained for a variety of tasks such as generating images from sketches or text, changing a photo of a horse into a zebra, or creating a whole new image of someone/thing that never existed. Using this technology, I trained a GAN using a dataset of cover art from Science Fiction works to see what landscapes, themes, or monstrosities might be created. While GANs create new art, trained models reveal familiar aspects from the original dataset like popular colors, shapes, and compositions. They also recreate and mimic the imaginative landscapes and worlds from the original covers meant to capture the eyes of the consumer. Specifically, using covers from Science Fiction merges overarching themes of the genre like technology, futurity, exploration, with contemporary digital methods as well as the colonialist, racist, and misogynist practices in each.

My presentation will cover my project’s workflow discussing the data collection, cleaning, and training process to create a trained model that generates new cover ‘art’ from a dataset of Science Fiction covers and the creative and ethical concerns encountered throughout the process.