5 Conclusion
The following findings can be drawn from the research about generative AI models: 1) Advancements and Limitations of GANs: Generative Adversarial Networks (GANs) have laid a crucial foundation in the field of image generation. But issues such as mode collapse limit their application, necessitating future improvements in stability and diversity during generation. 2) Impact of Diffusion Models: Diffusion Models (DMs), especially Latent Diffusion Models (LDMs), have significantly enhanced the quality and efficiency of image generation by reducing the dimensionality of the data, thereby ensuring high-fidelity output. 3) Role of Foundation Models in Generative AI: Foundation models have demonstrated exceptional generalization capabilities in natural language processing (NLP) and computer vision, particularly in unsupervised and transfer learning, facilitating their broad application across multiple fields of generative AI. As generative AI models are increasingly applied to more complex tasks, such as video and 3D modelling generation, these areas are expected to become key directions for future research in generative AI models.
The following findings can be drawn from the research about generative AI for different architectural design steps: 1) Evolution of Generative AI Applications: generative AI applications in architectural design have evolved through three stages. Initially, the focus was on image generation. This progressed to creating images with 3D information. Currently, exploration includes generating diverse images, 3D models, and videos. This evolution has optimized the design steps, significantly improving efficiency in tasks such as conceptual image generation, 3D modelling, and floor plan design. Architects can now quickly generate a variety of creative designs. 2) Future Directions of Generative AI Applications: Advancements in technology are expected to enable generative AI to enhance personalization and real-time editing in architectural design. Specifically, developing Architectural 3D and image generative models will allow designers to make immediate adjustments based on user requirements. Future research may focus on integrating automatic performance optimization that considers building regulations and environmental standards, allowing designs to balance creative expression with functionality and sustainability.
This paper reviewed the advancements and applications of generative AI in architectural design. The following are the main contributions of this paper: 1) Provided a quick overview of generative AI technology development—this paper summarized the development of generative AI technologies, encompassing various visual and language generation models such as GANs, VAEs, visual large models, and LLMs. 2) Summarized the applications of generative AI in architectural design—this paper systematically organized and analyzed the application methods of generative AI in various architectural design tasks, including conceptual design, 3D form design, floor plan design, facade design, and structural design. A detailed review of relevant literature enables readers to quickly understand the current state of generative AI applications in the architectural design process. 3) Utilized case studies to illustrate the methods of applying AI in architecture—this paper demonstrated the innovative and practical value of generative AI in architectural design through specific application examples. It discussed in detail the different technology application scenarios in architectural design. These examples encompass various aspects such as buildings, urban blocks, and interior spaces, thereby providing readers with insights and inspiration for practical applications. 4) Predicted future directions and challenges—this paper proposed potential future research directions such as multidisciplinary integration, user-participatory interactive design, and the application of emerging generative AI models, thus providing reference points for further research. Moreover, this paper predicted the main challenges in the future application of generative AI in architectural design, including data acquisition and processing, model complexity, and professional barriers. Overall, this paper explored how integrating generative AI models into architectural design enhances schemes, simplifies processes, and boosts efficiency.
Despite its contributions, this study has limitations that should be considered when interpreting the findings. The literature reviewed may not be fully comprehensive due to the random sampling of journals and conferences, potentially overlooking relevant studies. The research relies on existing literature, which may require updates as technology advances. The application of generative AI in section design is limited. Current AI technologies focus on specific design steps rather than the full architectural design process. Future research should explore more comprehensive AI applications across the entire design workflow.