Gameplay video for world models and game-playing agents

World models and game-playing agents need huge volumes of gameplay video with consistent action and consequence. Public game footage is a rights minefield, and synthetic-only data carries a domain gap — labs need licensed, diverse, real human play across titles and genres.

Our streamer network and connected catalogues turn real gameplay into a licensed corpus. We span genres — FPS, driving, platformers, sandbox, strategy — and tag by title type, perspective and pace, so a lab can pull the slice that matches the dynamics it wants to model.

Gameplay is a uniquely rich training signal — controllable, physics-consistent, endlessly varied. Teams use it for world models, sim-to-real RL environments, video generation, and agents that learn to act from observation — with a clean rights position the open web cannot match.