This paper presents a novel AI-aided architectural design method driven by daylight, aiming to generate preliminary architectural designs from massing models. The method integrates three main components: massing model generation, daylight-driven façade design, and architectural design generation using large-scale language models and text-to-image models. 1. **Massing Model Generation**: The method uses a parametric algorithm to generate a diverse array of massing models by adding and subtracting volumes, inspired by traditional architectural design strategies. This process is facilitated using 3D modeling software and Grasshopper. 2. **Daylight-Driven Strategy**: To address the lack of daylighting data in architectural design, the authors construct a dedicated dataset of daylighting maps. They use these maps to optimize façade designs by accurately determining window layouts. This involves training a LoRA (Low-Rank Adaptation) model with the collected floorplans and daylight parameters. 3. **Architectural Design Generation**: The final step involves using a large-scale language model (GPT-4) to generate text prompts based on architectural terms, which are then fed into a Stable Diffusion v1.5 model. Additional controls are integrated using ControlNet to ensure the generated designs align with the massing models. Experimental results demonstrate that the proposed method efficiently supports architects in conceptualizing and creating initial design proposals, enhancing design efficiency and creativity. User studies confirm the practical applicability and effectiveness of the approach. Future work aims to extend the method to more architectural physical factors and validate it across a broader range of architectural typologies.This paper presents a novel AI-aided architectural design method driven by daylight, aiming to generate preliminary architectural designs from massing models. The method integrates three main components: massing model generation, daylight-driven façade design, and architectural design generation using large-scale language models and text-to-image models. 1. **Massing Model Generation**: The method uses a parametric algorithm to generate a diverse array of massing models by adding and subtracting volumes, inspired by traditional architectural design strategies. This process is facilitated using 3D modeling software and Grasshopper. 2. **Daylight-Driven Strategy**: To address the lack of daylighting data in architectural design, the authors construct a dedicated dataset of daylighting maps. They use these maps to optimize façade designs by accurately determining window layouts. This involves training a LoRA (Low-Rank Adaptation) model with the collected floorplans and daylight parameters. 3. **Architectural Design Generation**: The final step involves using a large-scale language model (GPT-4) to generate text prompts based on architectural terms, which are then fed into a Stable Diffusion v1.5 model. Additional controls are integrated using ControlNet to ensure the generated designs align with the massing models. Experimental results demonstrate that the proposed method efficiently supports architects in conceptualizing and creating initial design proposals, enhancing design efficiency and creativity. User studies confirm the practical applicability and effectiveness of the approach. Future work aims to extend the method to more architectural physical factors and validate it across a broader range of architectural typologies.