Refik Anadol

Spatial Capture as AI Training Data

Refik Anadol is one of the world’s foremost media artists, known for large-scale AI-driven installations that transform raw data into immersive visual experiences. Visualskies contributed 3D scan data as training material for one of his nature-themed works—providing high-fidelity spatial datasets that informed the machine learning models at the heart of the piece.

The project required the capture of natural environments in precise three-dimensional detail: terrain, organic forms and landscape features recorded using a combination of terrestrial LiDAR and photogrammetry. The resulting datasets—point clouds with full colour capture—offered the richness of real-world geometry that distinguishes scan-derived AI training data from photographically assembled alternatives.

Working at the intersection of spatial capture and generative AI, this project represents a growing area of Visualskies’ work: providing ethically sourced, purpose-built scan datasets for AI systems in art, film and commercial applications. Where generative models are trained on scan data rather than scraped imagery, the outputs carry a spatial coherence and material authenticity that purely photographic training sets cannot match.

The collaboration with Refik Anadol’s studio demonstrated the creative potential of professional 3D capture as an input to machine learning—and the particular value of scan data in producing AI art that feels genuinely rooted in the physical world.