Generated Stereotypes

This project investigates biases and stereotypes embedded in generative AI image models by analyzing their outputs for culturally and politically loaded labels such as “prisoner,” “police officer,” “liberal,” “conservative,” “immigrant,” and “I.C.E. officer.” Utilizing generative AI platforms (Dalle, Flux and MidJourney), I prompt each model to produce 100 images per label. These outputs are then combined using pixel averaging techniques to generate a single, composite portrait representing the “average” image for each category. Through these synthesized portraits, the project reveals implicit assumptions, visual stereotypes, and underlying biases entrenched in AI image-generation processes.