An exploration of gender representation in generative AI images
Generative AI models have taken the internet by storm. They are increasingly used in many contexts, from stock photography to illustration. These models are used both to create new images and to enhance existing images. A user of Adobe Photoshop can remove an undesirable portion of a real photograph and replace it with generative content. Often, the blend between the real and generative portions of a photograph are hard to discern.
What assumptions are these models making? What biases are they learning? What are the implications of these biases? These are big questions that warrant years of research. This project analyzes the representation of gender in Stable Diffusion, a wildly popular open source generative AI model.