Immanuel Koh · CLIENT: Immanuel Koh
Singapore is an artificial city. Trained with a large dataset of 3D digital models of high-rise buildings found in Singapore, the custom designed AI model generates not only formally plausible and semantically coherent configurations, but begins to also imagine novel and uncanny architectural forms, interpolating and extrapolating among standard high-rise housing typologies such as the slab, point, and cluster blocks.
Press Kit
- Unique Properties
- Is it possible to sample at scale and at high-resolution every single building ever built in a country? If so, can such a dataset be used to train an AI model in generating new yet locally compliant buildings without any explicit regulatory control inputs? The project explores the design agency of deep generative neural networks in learning three-dimensional exteriority and interiority with a redesigned 3D generative adversarial network (3D-GAN) AI model. Winner of WAFX2024, IDA2024 and BLT2024.
- Dimensions & Packaging
- This is a digital AI x Architecture project. Each generated output is a new high-rise high-density residential public housing of up to 30 floors.
- Tags
- AI
- Production Technology
- A custom generative AI model was trained from scratch using proprietary code base and dataset. Specifically, a 3D generative adversarial network (3D-GAN) AI model was trained to learn semantic and spatial configuration of high-rise high-density residential public housing in Singapore.
- Design Challenge
- The non-existence of 3D models and their corresponding floor plans was the first challenge. New workflow and tools had to be developed to not only collect (i.e., digitally and physically), but create (annotate) and curate (i.e., exploratory data analysis) such high quality dataset to train an AI model from scratch. Technically difficult due to the scalability of similar CAD representation which was overcome via a novel AI latent representation of 3D.
- Project Duration
- This is an on-going and longer-term AI x Architecture research being conducted at the university design lab since 2021. It was however only published much later as an official architectural design award entry from 2024 onwards.
- Operation Flow
- This is a digital AI x Architecture generative research project, while the dataset was digitally collected from existing physical buildings in Singapore.
- Research
- A custom generative AI model was trained from scratch using proprietary code base and dataset. Specifically, a 3D generative adversarial network (3D-GAN) AI model was trained to learn semantic and spatial configuration of high-rise high-density residential public housing in Singapore.
- Inspiration
- Singapore is an artificial city! Trained with a large dataset of 3D digital models of high-rise buildings found in Singapore, the custom-designed AI model generates not only formally plausible and semantically coherent configurations, but begins to also imagine novel and uncanny architectural forms, interpolating and extrapolating among standard high-rise housing typologies such as the slab, point, and cluster blocks. Project exhibited at Venice Architecture Biennale and Singapore's Arts House.
- Image Credits
- All credits: Artificial-Architecture
- Patents
- Currently no patent.
Press releases are available in 1 languages.