Automated facade solar retrofits using feed-forward 3D reconstruction and generative image editing

Building-integrated photovoltaics (BIPV) enable solar energy generation from facades, yet facade retrofit assessment remains largely manual and time-consuming. This paper introduces a pipeline for automated analysis and visualization of BIPV retrofit designs. Unlike prior studies relying on building footprints, 3D models, or camera parameters, the proposed method automatically estimates facade aspect ratios and dimensions from images using feed-forward 3D reconstruction and vision-language models. The pipeline edits facade images using pretrained generative models to visualize BIPV retrofits considering system size and provides preliminary energy-yield estimates. On two datasets covering facades from 10 cities, the best-performing configurations achieved normalized mean absolute errors of 9% and 15% in aspect ratio estimation and an average floor-count prediction error of 0.5 floors. The pipeline enables scalable evaluation of early-stage facade solar retrofit designs, supporting decision-making in retrofit planning by reducing reliance on manual surveying and modeling. The pipeline implementation and dataset are available to support future work.