Automated residential layout generation and editing using natural language and images

Architectural design, including for the most common residential buildings, is a complex process that typically requires iterative revisions by skilled architects. This paper addresses how to automate the generation and modification of residential layouts, to lower the design threshold and enable cost-effective, user-driven generation and editing. This paper proposes Text2FloorEdit, a framework that decomposes the design task into three components: Residential Layout Generation (RL-Net) for flexible residential layout generation; Window, Door, and Wall Generation (WD-Net) for detailed floor plan generation with lower training costs; and a 3D rendering system for visualisation. The proposed approach enables the efficient generation and modification of residential layouts using flexible inputs like natural language and images, without the need for multimodal datasets. This solution is particularly valuable for architects and non-professionals seeking cost-effective, user-friendly tools for automated residential design. This paper opens new directions in cross-modal generative models, with the potential to enhance architectural design automation.