Licensed gameplay datasets for world models and embodied AI.

We source licensed games, capture gameplay and build structured datasets for AI training and evaluation.

record / frame 812illustrative, fields vary by title
{
  "session": "s_0142",
  "frame": 812,  "t_ms": 27066,
  "rgb": "frames/000812.png",
  "camera": { "pos": [3.92, 1.60, 16.04], "yaw": 154 },
  "input": { "keys": ["W", "D"], "mouse_dx": 4 },
  "action": "push",
  "event": "contact:crate_3",
  "depth": "depth/000812.exr"
}
01Capture pipeline

From licensed game to training-ready dataset.

  1. 01

    Verify rights

    Secure and verify the rights required to capture and license gameplay data for AI training.

  2. 02

    Record gameplay

    Capture gameplay sessions from licensed games using the OneTwoPlAI capture pipeline.

  3. 03

    Create dataset

    Structure, align, enrich and package captured gameplay into training-ready datasets.

02Dataset source

Games.

Licensed titles, captured in play

Supply is organised by genre, by the interactions a model needs to learn, and by the environments inside the game.

By genre

  • FPS
  • Action
  • RPG
  • Open world
  • Racing
  • Simulation
  • Survival
  • Co-op
  • Strategy

By interaction and mechanics

  • Navigation
  • Combat
  • Driving
  • Object interaction
  • Puzzle solving
  • Resource management
  • Cooperative actions

By environment

  • Urban
  • Natural
  • Industrial
  • Interior
  • Fantasy
  • Sci-fi

Environments as they appear inside licensed games, captured during play. Not standalone 3D assets.

03What we capture

Synchronised gameplay, inputs and labels.

RGB gameplay

The rendered game at capture resolution and frame rate.

Player inputs

Keyboard, mouse and controller input with timing.

Camera movement

Camera position and rotation over time.

Actions and events

What the player did and what happened in the game.

Movement and trajectory

The path through the world, frame by frame.

Interaction sequences

Multi-step interactions with objects, characters and systems.

Temporal alignment

Every stream on one clock, aligned to the frame.

Labels and metadata

Session, title, task and scene metadata.

Depth and normalswhere available

Per-frame depth and surface normals, depending on the title.

04Who we serve

Built for models that need interaction, not just video.

AI foundation labs
World-model teams
Robotics and embodied AI
Game-AI companies
Video-generation teams
Planning and evaluation teams
Sim-to-real research
05Custom capture

Need something specific?

Tell us the genre, mechanics, environment, duration and modalities you need. We can source suitable licensed titles and capture to specification.

Contact

Discuss data requirements.

Share what you are training and what the data needs to contain. We will come back with suitable titles and a capture plan.

Required modalities