Generative artificial intelligence in building facade design: Models, applications, and future directions

Source abstract

Facade systems in buildings contribute significantly to buildings’ aesthetic qualities, performance, and occupants’ comfort. Recent advancements in generative artificial intelligence (AI) have created new opportunities for building design exploration. Despite the growing applications of generative AI in built environment research, a focused review study on its use in building facade research and design remains unexplored. Through content analysis and a critical review approach, this study examines the research landscape of generative AI in building facade design, with emphasis on model categories, data types, evaluation strategies, and application domains. The findings show that generative adversarial networks (GANs) remain the most widely adopted models in AI-related facade research. Also, image-based data type dominates both model input and output, while visual assessment and Structural Similarity Index (SSIM) are the most frequently used evaluation methods. Furthermore, seven major application domains were identified, including facade restoration and reconstruction, design and visualization, sustainability-oriented investigations, human-centered evaluation, building and urban attribute analysis, facade deterioration and damage forecasting, as well as heritage and historic facade generation. Finally, the study identified several key research gaps, including the under exploration of autoencoder-based models, NeRF, and Gaussian Splatting, the need for more integration of generative AI with facade systems research and performance-driven design workflows, the development of architecture-specific generative AI tools, the limited adoption of geometry-aware and immersive workflows, and the underdeveloped area of human-centered facade evaluation. Overall, the review provides a foundation for advancing performance-oriented generative AI applications in facade design.