CATEGORY // WORLD MODEL

Player–environment interactions for world-model training.

Overview Production
CATEGORYWorld ModelIN STOCK8,000+ hoursCAPACITY10,000 hours / weekSPEC≥1080P / ≥60FPS

01 / DATASET OVERVIEW

MULTIMODAL DATA

SCOPE & PURPOSE

The training task.

  • GAMEPLAY VIDEO
  • ACTION DATA
  • WORLD MODELS

Starting from gameplay video, captures player–environment interaction data via low-level reverse engineering of games, producing "UI-free, highly synchronized, ground-truth-action" world-model training data.

02 / DATASET PROFILE & TECHNICAL SIGNALS

SCOPE · SCALE · CONSTRUCTION
01IN STOCK
8,000+ hours
02CAPACITY
10,000 hours / week
03SPEC
≥1080P / ≥60FPS
04TRAINING SIGNAL
Action-aligned
Ground-truth player–environment interactions.

03 / DATA PRODUCTION & CONSTRUCTION

METHOD & WORKFLOW
  1. 01SOURCE

    Gameplay video

    Start with gameplay video to capture player–environment interactions.

  2. 02EXTRACT

    Low-level reverse engineering

    Capture interaction data through low-level reverse engineering of games.

  3. 03STRUCTURE

    Highly synchronized data

    Synchronize the visual signal with ground-truth action data.

  4. 04OUTPUT

    World-model training data

    Produce UI-free, highly synchronized, ground-truth-action data.

04 / COVERAGE & CAPABILITY MATRIX

TASKS · METHODS · APPLICATIONS
DIMENSIONCOVERAGE
01Source environment
Gameplay videoPlayer–environment interactions
02Training signals
UI-free videoGround-truth action
03Data construction
Low-level reverse engineeringHigh synchronization
04Model application
World-model training
Explore the production method

NEXT STEP