← Back to Computer Vision
cs.CV

Do artistic styles unconsciously encode gender bias?

Piera Riccio, Miriam Doh, Benedikt Höltgen, Noa Garcia, Nanne van Noord

June 4, 2026

Artistic styles aren't neutral filters—they encode gendered visual conventions rooted in historical social hierarchies. Researchers created StyleGender, a dataset linking 19 art historical styles to gender representation, and measured how these patterns transfer to AI-generated images. Text-to-image models don't just copy art history; they amplify gender artifacts beyond what appears in original paintings, suggesting generative AI can magnify latent biases in training data.
Published as Gender Artifacts from Art History to Text-to-Image Generation arXiv:2606.05829
Read the original paper →