In recent years, the rapid development of large-scale models has made new possibilities for interdisciplinary fields such as architecture. In this paper, we present a novel daylight-driven AI-aided architectural design method. Firstly, we formulate a method for generating massing models, producing architectural massing models using random parameters quickly. Subsequently, we integrate a daylight-driven façade design strategy, accurately determining window layouts and applying them to the massing models. Finally, we seamlessly combine a large-scale language model with a text-to-image model, enhancing the efficiency of generating visual architectural design renderings. Experimental results demonstrate that our approach supports architects' creative inspirations and pioneers novel avenues for architectural design development.
Our pipeline shows an AI-aided architectural design workflow. Massing Generation (MG) is the module for generating architectural massing models, DDS is the daylight-driven strategy, and Architectural Design (ADG) is the module utilizing large-scale models to generate architectural designs.
we present a façade generation approach centered around daylighting. Given the lack of readily available daylighting data, we undertake a series of steps to construct a dedicated dataset of daylighting maps for training the LoRA (Low-Rank Adaptation) model. We then input representative sectional profiles of massing models into the trained model to generate corresponding daylighting maps. This process accurately determines window layouts and precisely integrates them onto the massing models.
We employ random parameters to generate massing models, serving as foundational elements for architectural design. As illustrated in Figure 5, our approach initially extracts three sectional profiles from contour lines. We then generate daylighting maps using the LoRA model, filtering them meticulously to ensure plausible architectural façade designs. Based on these filtered maps, we create façade window profiles. Leveraging diverse architectural text prompts, we generate distinctive design proposals showcased in (d). These proposals excel in exterior proportions and visual renderings, providing architects with valuable support for realizing creative concepts efficiently.
Refer to the pdf paper linked above for more details on qualitative, quantitative, and ablation studies.
We propose a novel AI-aided architectural design method. We first generate numerous massing models as shown in (a).
Then, a specific massing model (b) is selected. Following this, guided by a daylight-driven strategy, we refine the architectural façade depicted in (d), achieving efficient generations of the architectural design in (e) by using diffusion models.
@incollection{li2023layerdiffusion,
title={Generating Daylight-driven Architectural Design via Diffusion Models},
author={Li, Pengzhi and Li, Baijuan},
booktitle={arXiv preprint arXiv:2404.13353},
year={2024}
}