When Culture Becomes Costume: How Africans Interpret and Contest AI-Generated Traditional Attire

Abstract

Recent developments in generative AI enable the rapid production of lifelike images, but these tools often reproduce cultural stereotypes for underrepresented groups. We adopt a cross-regional, qualitative-dominant exploratory design to examine how text-to-image (T2I) systems portray African identities and how community members interpret these portrayals. The study combines a controlled image-generation task with semi-structured interviews with 18 participants from North and West Africa (Nigeria, Tunisia, Egypt, and Morocco).. Participants generated images of everyday portraits, traditional attire during celebration, a professional context, and a speculative future depiction. Focusing on “traditional attire”, we identify three recurring interpretive tensions participants used to diagnose cultural misrepresentation: accessories (e.g., beads) functioning as shallow cultural proof, garment-semantics collapse when textiles, cuts, and named garments are missing, and temporal collapse in which living tradition is rendered as museum-era costume. We show how participants draw on garment knowledge and temporal reasoning to assess authenticity, and we argue that participants’ accounts suggest that culturally situated representation cannot be reduced to prompt-level debiasing alone.