High fashion archives, artificial intelligence, and the homogenization of the ideal body

Abstract

The present article analyzes the role of digital high fashion archives in the training of generative artificial intelligence (AI) models and examines how these datasets contribute to the reproduction and the reinforcement of a normative body ideal. Far from being neutral repositories, these archives actively participate in the consolidation of a scopic regime that privileges specific physical traits—extreme thinness, youth, and whiteness—at the expense of bodily diversity. Drawing on an interdisciplinary approach that integrates fashion studies, critical technology studies, and visual studies, the article examines the relationship between the historical curation of fashion archives and contemporary algorithmic biases. The study is conducted in two phases: a visual genealogy of the body ideal in twentieth-century fashion photography, and an empirical analysis of images generated by AI models such as Midjourney and Stable Diffusion. The findings show that these systems not only replicate archival standards but also intensify them through processes of algorithmic homogenization. The article argues that the apparent objectivity of AI reinforces exclusionary dynamics of bodily representation, with significant social, cultural, and political implications, and calls for more critical and responsible practices of data curation and algorithmic design.