Generating Daylight-driven Architectural Design via Diffusion Models

TL;DR

The paper tackles the challenge of integrating daylight considerations into AI-driven architectural ideation by presenting an end-to-end pipeline that moves from random massing generation to daylight-aware façade design and rendered visuals. It combines Massing Model Generation (MG), a daylight-driven strategy (DDS) for window placement, and Architectural Design Generation (ADG) that fuses GPT-4 prompted design with diffusion models under ControlNet and LoRA conditioning, guided by a newly constructed daylighting dataset. The work provides practical contributions: a parametric massing workflow, a daylight-informed generation framework, and an integrated text-to-image rendering approach validated by a user study with practicing architects. Overall, the approach accelerates early-stage design and promotes daylight-conscious architectural outcomes, with potential to influence daylight-focused workflows in practice.

Abstract

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 facade 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. Project page: https://zrealli.github.io/DDADesign/.

Figures (7)

• Figure 1: 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.
• Figure 2: We demonstrate the processes of addition and subtraction in the generation of massing models. The generation of new massing models is founded upon these fundamental operations.
• Figure 3: We illustrate the process of generating our daylighting data. First, the floorplan (a) is decomposed into three vector components (b): interior walls, exterior walls, and windows (red). Subsequently, daylight parameters and algorithms are employed to compute the corresponding daylighting maps.
• Figure 4: More architectural designs. (a) is the generated massing model, (b) illustrates the daylighting map corresponding to (a). (c) shows the massing model following façade optimization, and (d) displays the visual renderings of the architectural design. '#' divides the text prompts into four corresponding categories.
• Figure 5: We present more results of daylighting data generation. The upper figure showcases partial results of daylighting maps, while the lower figure shows some collected floorplans.