Game data
for world models.
Quality-filtered multiplayer capture — RGB, segmentation, depth, normal, motion vector, and full telemetry — ready to train on.
Most available game footage shows what a world looks like. Not how it behaves.
World models need to learn motion, physics, spatial structure, and cause-and-effect — signals that a single RGB frame can't carry. Licensed game engines already compute that information internally. Layer Labs captures it at the source, filters it for quality, and delivers it in the structured form research teams actually train on.
Eight synchronized layers, one capture pass
Every session is captured once and delivered as a full, time-aligned stack — not stitched together after the fact.
RGB
Full-resolution gameplay video, filtered for quality and content before it enters any dataset.
Semantic segmentation
Per-frame object and surface masks, aligned to the RGB pass for scene understanding.
Depth
Dense depth buffers giving models real spatial structure, not inferred geometry.
Normal
Surface-orientation maps for accurate lighting, shading, and geometry reasoning.
Motion vector
Per-pixel motion between frames — the signal world models need to learn how things move.
Camera telemetry
Exact camera pose, FOV, and movement, frame-synced to every other layer.
Player telemetry
Inputs, actions, and state from real multiplayer sessions — genuine human behavior, not scripted.
Vehicle telemetry
Physics-grounded position, velocity, and control data from in-game vehicles.
Built to ship at dataset scale
Existing catalogs ready to license, plus custom capture built to your spec.
Built by people who shipped the games this data comes from.
Layer Labs is founded by industry veterans from PlayStation, Xbox, and Unity — teams that understand game engines, capture pipelines, and what "production-ready" actually means to a research buyer.
Filtered, not scraped
Every session is reviewed for quality before it enters a dataset — we don't deliver raw, unfiltered capture.
Captured at the engine, not the screen
Depth, normal, motion vector, and telemetry come from inside the pipeline — not reconstructed after the fact.
Real multiplayer behavior
Player and vehicle telemetry comes from genuine multiplayer sessions, not scripted or synthetic playthroughs.
Tell us what your model needs to see.
Whether you're evaluating an existing catalog or scoping a custom capture, send us a few details and we'll follow up with sample data and specs.