Gameplay data for multiplayer world models.

Realis Worlds is building a gaming capture system for world model focused AI labs. We reward gamers to play in squads, capture their sessions, and turn that activity into gameplay video paired with action labels.

Labs use resulting datasets to train the frontier of AI models.

The focus is the shared timeline: what each player saw, what each player did, and how those actions interacted across the same event.

Inquiries
Backed by researchers at
National University of Singapore logo
University of Toronto logo
University of Waterloo logo
Western University logo
Overview

Collection and delivery.

Realis rewards gaming squads to play together, captures each player stream with action labels, and delivers synchronized multiplayer session records to AI labs.

Source sessions
Most data comes from rewarded squads playing live multiplayer sessions. When a lab needs a particular game or clip type, we can post targeted bounties.
Capture behavior
Each player is recorded separately with gameplay video and action labels, preserving the same event across multiple viewpoints in a shared timeline.
Deliver datasets
Labs receive synchronized session records for training and evaluation in multi-POV world modeling, coordination, action prediction, and long-horizon interaction.
Contact

For labs and gaming squads.

If you want access to the data, or you want to participate in collection, reach out.

contact@realisworlds.com