This is the authors' abstract. We don't add key points for this paper.
Understanding how design students evaluate architectural visual cues is complicated by real-world building variation. Generative AI can produce comparable facade alternatives from a common baseline, though it transforms multiple visual attributes simultaneously rather than isolating pure “style”. This study examines visual preference and decision difficulty among spatial design students in Taiwan using AI-generated transformations based on seven architect-style prompt categories applied to familiar campus buildings. A cohort of 106 students completed a pairwise forced-choice task, yielding 3,298 valid judgments. To address the methodological challenges of comparative research, preferences were analyzed using a Bradley–Terry-style logistic model, and the AI stimuli were validated through an independent expert recognition task, which confirmed high category recognizability (81.3% accuracy) despite geometric drift. The model revealed a structured preference hierarchy, with the Zaha Hadid, Richard Meier, and Oscar Niemeyer categories receiving the highest estimated utilities. Furthermore, mixed-model analysis of reaction times demonstrated that decisions were significantly faster when the overall preference gap between the two architectural options was larger. While adopting a cautious approach to claims of pure style isolation, this study contributes to architectural education by offering a rigorous, audited framework for using generative AI imagery to investigate visual judgment and design cognition.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: