Multi-agent collaborative pathways for Chinese traditional architectural image generation

• Published in: Scientific reports Vol. 15; no. 1; pp. 34596 - 15
• Main Authors: Lu, Yi, Yuan, Wenmiao, Wang, Mengyao, Wang, Pohsun, Wu, Sifan, Wu, Jiacheng, Xing, Wenzhuo, Xie, Wanxu, Yu, Fuze
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 03.10.2025; Nature Publishing Group; Nature Portfolio
• Subjects: 639/166; 639/705; 704/172; 704/844; Architecture; Artificial intelligence; Chinese traditional architecture; Collaboration; Cultural heritage; Cultural tourism; Digital cultural tourism; Generative artificial intelligence; Humanities and Social Sciences; Multi-agent collaboration; multidisciplinary; Science; Science (multidisciplinary); Tourism; Beijing China; China; Multi-agent collaboration; Generative artificial intelligence; Chinese traditional architecture; Digital cultural tourism
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/s41598-025-18130-7

Artificial Intelligence Generated Content (AIGC) technology demonstrates significant potential in the fields of cultural heritage digitalization and cultural tourism design. However, when confronted with specific subjects such as Chinese traditional architecture, which embodies rich cultural connotations and complex visual elements, existing technologies still exhibit limitations in understanding vague user requirements, ensuring cultural accuracy, and generating diverse and high-fidelity images. To address these challenges, this study proposes a text-to-image generation framework oriented towards Chinese traditional architecture, based on multi-agent collaboration. This framework integrates multiple intelligent agents responsible for user intent understanding, creative prompt generation, traditional architectural image generation, aesthetic and cultural relevance assessment, and collaborative workflow scheduling. By constructing a Chinese Traditional Architecture Cultural Knowledge Base (exemplified by the Beijing Central Axis) and designing a collaborative workflow among the agents, the framework can efficiently and accurately transform users’ colloquial and vague descriptions into Chinese traditional architectural images with profound cultural connotations and high visual fidelity. Experimental results demonstrate that this collaborative multi-agent framework significantly outperforms baseline models in terms of handling vague inputs, ensuring the cultural accuracy of generated images, enhancing user intent matching, and creative diversity. This research not only provides new theoretical perspectives and practical pathways for AI technology in the digital preservation, creative transformation, and intelligent design of cultural heritage, but also offers an effective tool for empowering innovative digital cultural tourism experiences, holding significance for technological exploration and cultural inheritance.