3D Neural Synthesis demonstrates the application of generative AI model to explore participatory models of designing low-carbon futures. Building on previous research in 3D Generative AI, the project empowers designers and stakeholders to engage with sustainable design solutions through an innovative combination of AI and architectural tools.

‍

Initially presented at the 2023 ACADIA conference in Denver, the research introduced a pioneering approach for redesigning 3D models using text prompts. The process involves 3D modeling in Rhino, extracting camera frames, and applying generative image-to-image models guided by text inputs. The resulting images are then reconstructed in 3D using Neural Radiance Fields (NeRF), allowing for new creative possibilities in design visualization.

‍

Since the paper’s publication, the workflow has been further refined with the integration of 3D Gaussian splats and Unity workflows, combined with the Google Maps API. This enhancement enables users to digitally "scan" a site, apply modifications using text prompts and image processors, and visualize new sustainable building designs in real-time. This approach not only advances creative exploration but also reinforces the importance of sustainability in future architectural practices.

View the Paper Link here