From Concept to Consistent Multi-View Renders: Integrating shape grammars with Stable Diffusion in architectural workflow

The incorporation of artificial intelligence (AI) in architectural design has progressed markedly. Nonetheless, current workflows are constrained. A major issue is the inconsistency in AIgenerated renderings from different perspectives of the same architectural design. Moreover, there is the problem of insufficient integration with the conceptual design phase. This paper introduces a unified workflow that combines shape grammars with Stable Diffusion to enhance architectural visualization. The proposed methodology employs a three-stage computational process: first, upon inputting the design parti, it produces diverse and feasible floorplans via shape grammars implemented in the Shape Machine; next, it extrudes these floorplans into three-dimensional models; finally, it renders these models into high-quality, cohesive interior visuals from various perspectives utilizing Stable Diffusion, enhanced with ControlNet and LoRA. The results demonstrate that our autonomous and efficient workflow substantially reduces design and rendering time, while enhancing control and flexibility in design generation. This workflow streamlines the design process from initial concepts to final presentations, improving the quality and consistency of renders relative to conventional AI workflows.