Abstract The proliferation of artificial intelligence (AI) image generation presents both opportunities and challenges for preserving fashion heritage and cultural identity. This study aims to evaluate AI’s capacity to replicate historical fashion by introducing the Multi-Dimensional Fidelity Assessment (MDFA) framework, which combines computational perceptual similarity (LPIPS) with expert-scored cultural feature concordance (CFC). Applying MDFA to 15 ensembles generated by Gemini and ChatGPT, findings reveal a divergence between automated metrics and human judgment. While AI often achieves high perceptual similarity, fidelity to culturally significant details is inconsistent. Consequently, these tools risk cultural flattening and misrepresentation, especially for non-Western attire. For example, AI struggled with symbolic motifs on a Qing Dynasty robe and Indonesian batik textures. The paper argues for ethical mediation in virtual archives by emphasizing transparency, contextualization, and expert oversight to ensure AI serves as a responsible tool in safeguarding diverse fashion legacies. In turn, the findings provide insights into identifying and addressing falsifications and outline directions for future research in digital heritage forensics.
Al-Ghamdi et al. (Sat,) studied this question.