HyperLayout: A decoupled and planning-oriented framework for parcel-level building layout generation

Urban development is increasingly shifting from outward expansion toward the renewal and optimization of existing urban areas. This transition has created a growing demand for building layout generation methods that can satisfy complex planning requirements while maintaining sound spatial organization. However, most existing end-to-end approaches jointly model building attributes, morphology, and spatial arrangement. This unified process makes it difficult to trace how planning requirements affect spatial outcomes, inspect intermediate decisions, or adjust generated scenarios. This study proposes HyperLayout, a decoupled and planning-oriented framework for parcel-level building layout generation. Rather than treating layout generation as a single synthesis task, HyperLayout organizes it into three interpretable stages: development-capacity configuration, building form generation, and spatial layout optimization. Through explicit constraint propagation, parcel-level planning objectives are progressively converted into building attributes, geometric conditions, and spatial optimization criteria. This structure makes intermediate decisions and outputs easier to inspect and revise throughout the generation process. Experiments show that HyperLayout improves upon the evaluated baselines in both building form generation and building layout optimization. The framework produces layouts with improved geometric consistency, spatial organization, adaptation to parcel boundaries, and compliance with planning constraints. It also supports the controlled exploration of alternative scenarios under different planning objectives at parcel and community scales. These findings suggest that workflow decomposition and explicit constraint propagation provide a practical framework for developing interpretable and controllable AI-assisted planning tools for urban design and redevelopment.