The Visions dataset was built from over 10,000 public-domain artworks, later refined to only 666. These images were drawn from major open-access museum archives such as The Met, Rijksmuseum, NGA, Getty, AIC, and Europeana. Search terms that guided the curation included allegory, vanitas, trompe-l’œil, plague, apocalypse, mirror, mask, dream, and satire, revealing coded critiques beneath visual beauty and merging diverse art movements to form a dataset where seduction and critique coexist, providing the aesthetic and conceptual foundation for Visions.