A Structured Prompt Framework for AI-Generated Biophilic Architectural Spaces

This study proposes a structured prompt framework to improve the quality and consistency of generative visualizations in biophilic architectural space (BAS) design. Existing generative artificial intelligence (Gen AI)-based visualization approaches often lack alignment with biophilic design principles, resulting in outputs that fail to reflect the restorative qualities of nature-integrated spaces. To address this limitation, the study links generative visualization processes to established biophilic frameworks, thereby enhancing the applicability of Gen AI in sustainable and human-centered architectural practice. The methodology follows a three-stage process: (1) exploration of biophilic visualization requirements through literature review and prompt testing, (2) development of the framework through domain-specific dataset construction, text mining, and prompt curation, and (3) expert evaluation of images generated using the structured prompts. The proposed framework consists of five components—subject, attribute, mood, time and background, and negative prompt—to guide the generation of BAS visualizations systematically. The generated images were assessed based on five criteria: domain fidelity, visual coherence, depth and perspective, spatial integration, and overall biophilic appeal. Results demonstrated substantial improvements—up to 75% in domain fidelity and over 60% in spatial integration and biophilic appeal—compared to early-tested prompts. These findings underscore the potential of structured prompts, grounded in biophilic design theory, to enhance the effectiveness of AI visualizations. This study offers a replicable and scalable method for integrating nature-based design principles into early-stage spatial planning. It provides design professionals with a practical tool to visualize restorative environments and promote sustainable architectural practice.