World Labs Engine-Ready Scene Authoring

World Labs is positioning itself less like an AI camera and more like a scene-authoring system for production and simulation. Deterministic camera placement means a team can put the camera at the exact same coordinates every time for a game cutscene, VFX pass, or robotics benchmark. Collision meshes and asset-level control mean objects become usable building blocks inside Unity, Unreal, or simulators, instead of pixels that only look right from one moving viewpoint.

• Runway is strongest where immediacy matters. GWM Worlds 2 streams continuous 720p, 24 fps interactive video with subject and camera control, but Runway says difficult camera motion can degrade geometry and long-term memory is imperfect. That is workable for interactive storytelling, but weaker for pipelines that need stable 3D structure under repeatable camera paths.
  2 sacra 3 runway
• Marble is built around persistent 3D worlds that can be created, edited, shared, downloaded, and exported. That export step is the key difference. A game studio or robotics team can take the world out of the demo surface, inspect pieces, swap assets, attach physics, and run the same scene again under controlled conditions.
  1 sacra 4 worldlabs
• Google DeepMind shows the other branch of the market. Genie 3 generates navigable worlds at 720p and 24 fps and Google exposes it through Google AI Ultra in the U.S. That makes world models easier to try, but subscription distribution does not by itself solve the need for engine-ready geometry, collision proxies, and per-asset editing.
  5 deepmind 6 blog 7 immersion3d

The market is likely to split between video-native world models for instant interactive media and spatial world models for workflows that end in engines, simulators, and production tools. If World Labs keeps improving export fidelity and editing control, it can own the part of the stack where generated worlds stop being a demo and start becoming infrastructure.