Hair-SyleGAN

HairStyleGAN is an experimental encoding of text in algorithmically generated photographs of braided hair. The algorithm takes in a short to medium phrase as in input, and “trans - lates” it into a seed. Seeds are internal coordinates within a StyleGan2 vector that allows for the discrete location of a particulate section within the model. The resulting image encrypts a representation of a cultural artifact. In the images presented here, Fourth Author hides excerpts from Audre Lorde’s poem “A Woman Speaks” in views from the back of a person’s head. The stanza “I do not mix / love with pity / nor hate with scorn” appears below an image of several twisted coils meeting at the center. The braided head rests above an impossibly wilting neck, one of the only visible traces of algorithmic sorcery. To create the imagery, Fourth Author trained a model that would generate new braiding hairstyles, based on a dataset of unique braided hairstyles. “I trained a model for a total of 5,000 steps, and determined that it was effectively creating unique variations. After I achieved this I wanted to operationalize the model through a system, to revisit the function of language in traditional hair braiding technology.” These explorations grow from their reflections on Generative Fiction and how language models get built. Left in black and white, the synthetic photographs are sensual and arresting. They push the viewer to look closer rather than recoil from the erasure of Blackness, as Audre Lorde’s excerpt describes.