Methodology

Classification

• World creation: produces persistent or exportable 3D environments, panoramas, meshes, splats, or programmable world assets.
• Interactive worlds: generates an environment while a person or agent acts inside it, commonly through video, keyboard, controller, or text input.
• Spatial and physical AI: predicts geometry, physical change, or actions to support robotics, planning, mapping, and simulation.
• Research and adjacent systems: illuminates part of the world-model stack without providing a general explorable-world workflow.

These categories describe the public artifact, not a universal taxonomy. A system may span more than one category; it is placed where its current public capability is easiest to understand.

Source standard

Every entry must have at least one primary source from the maker: an official project page, product documentation, research paper, or maintained repository. Third-party reporting can help discovery, but it is not used to establish the entry’s status.

Access, pricing, performance, and release claims change quickly. The directory therefore records a verification date and avoids unsupported superlatives or benchmark comparisons.

Boundary notes

A generated video is not automatically an explorable world. A reconstructed scene is not automatically a physics simulator. A robot-control model is not a consumer world creator. Boundary notes make these differences explicit.

Corrections should include a primary source and the date it was checked. See the contribution guide.

Browse the directory or read the deeper profiles on World Models Watch.