Data / World Model Data
Data for models that understand how the world works.
World models learn physics, space, and cause and effect from observation. That requires large, well-annotated captures of the real world.
01 /Data types
What we source, generate, and evaluate.
01
Video
Diverse real-world video with consistent metadata.
02
Multimodal
Aligned video, audio, text, and sensor streams.
03
Spatial
Depth, 3D, and scene-level spatial data.
04
Physical-world observation
Captures of objects, motion, and environments over time.
05
Real-world interaction
People and agents acting on their surroundings.
06
Video-language
Video paired with dense, accurate language descriptions.
02 /How it is made
Every example carries a record of how it was made.
Source→Generate→Review→Evaluate→Exchange
This category is delivered through FelixSphere data engagements, with the same review and provenance standards used on UnifyData.
Built for
- World model and video model labs
- Autonomy and simulation teams
- Multimodal research groups
Compute · Data · Models · Intelligence
Tell us what your model needs to learn.
Describe the capability, the domain, and how you will measure it. We will scope the data.