Abstract

This study investigates the capability of generative artificial intelligence (GenAI) to produce architectural designs that align with specific stylistic movements, namely Postmodernism, Deconstructivism, and Biomimicry, based on prompts incorporating visual attributes such as beauty, color, proportion, and materials. Using a machine learningbased classification model, AI-generated images were compared against ground-truth examples representing each architectural style to assess both visual and conceptual alignment. The results revealed a notable disparity between the generated outputs and the authentic design principles of the intended styles. While some consistency was observed particularly for Postmodernism, most AI outputs failed to accurately capture the nuanced language and defining characteristics of the architectural movements. These findings underscore a broader limitation in current generative models, which often struggle to distinguish between visually and conceptually distinct styles when assessed through quantitative methods. The study recommends enriching the training datasets with a more diverse and balanced set of reference images representing various architectural traditions. Additionally, the integration of a qualitative evaluation process involving expert architects, designers, and students is proposed to provide a more context-sensitive assessment of stylistic accuracy. The study recommends that, by conjoining computational analysis with human expertise, a more comprehensive framework for evaluating AI-generated architectural designs will be offered, contributing to the ongoing discourse on the role of artificial intelligence in creative and design-driven fields.