Physicl

Rare moments are the product here. Physicl builds synthetic training data for physical AI, meaning robots, autonomous systems and world models, on the argument that the situations those systems fail in are almost impossible to film, label and replay. Instead of recording footage, a customer describes a scene and generates it from 3D worlds whose geometry and physics are already known.

Studio and its exact labels

Studio is the first app. A team states the situation its model gets wrong, then renders images and videos of it thousands of ways. Depth, surface normals, segmentation, boxes and camera pose come from the scene itself and arrive with every frame, so nobody has to annotate them afterwards. Lighting, layout, clutter, camera and objects can each be changed while everything else stays put, which lets two runs be compared directly. Physicl says the library holds 10,000 environments and 100,000 underlying assets built from real geometry with known materials. Studio is in beta, opened to research teams a few at a time and free during that period.

Robot policies and world models

The company's use cases centre on robot policies and world models, and its stated users are labs, universities and companies building vision, world model and robotics systems. It cites SpAItial, a maker of world models, as having more than halved its reconstruction error on real rooms after training on Physicl data.

A team from 3D production

Founder and chief executive Alex de Vigan spent about a decade digitising objects and rooms in 3D, including for large retailers. The site describes around 30 people in 3D, simulation and machine learning, most of whom worked together for six years before Physicl. Its advisers include Thomas Wolf of Hugging Face, Christoph Lassner of World Labs and Matthias Niessner of TUM. In November 2026 it announced a $10 million seed round led by Inovia Capital with US Venture Partners.

Key facts

Company: Physicl, Inc. Founder: Alex de Vigan, CEO Offices: New York, San Francisco and Paris Public launch: March 2026, at NVIDIA GTC Funding: $10 million seed, led by Inovia Capital (company statement, 6 November 2026) Product: Studio, in beta