Video World Models · 2026-10-09
WorldCast: Distributed Multiplayer World Models
Ziyang Ye, Junchao Huang, Evelyn Zhang, Zhihao Xie, Ruicheng Zhang, Boyao Han, Litao Ban, Ziye Wang, Xinting Hu, Shaoshuai Shi, Zhuotao Tian, Li Jiang
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
WorldCast is a distributed multiplayer world model: each player runs a local client with a video generator and a state model, and clients exchange only shared player and scene state. A camera-aligned player state field guides where other players render, while a shared scene-state memory bank lets clients reuse observations. On Counter-Strike 2, the camera-aligned player field improves player rendering rate by over an order of magnitude, and scene state improves whole-round visual consistency.
Why it matters
Maintaining cross-view consistency for independently controlled players without jointly generating all views is a central challenge. The authors propose a scalable approach where generation stays local and only lightweight world state is exchanged, aiming to reduce computation and inter-device communication growth as players increase.
Method
- Distributed clients: each player has a local video generator conditioned on shared world state; clients exchange only player state updates and scene-state memory entries once per block.
- Camera-aligned player state field: other players’ world-coordinate states are projected into each client’s token grid to provide spatial guidance (coverage, depth, facing, and gameplay attributes).
- Shared scene state memory bank: clients retrieve at most one memory entry to reuse generated observations for missing content, and a state model estimates position and depth from generated video and controls at inference.
Limitation
The evaluations use recorded first frames and GT/recorded controls; in whole-round results, the upper part of Table 19 and related results use GT states to separate generator/scene-state effects from predicted-state effects, and rollouts depart from the recording as errors accumulate. The state model predicts only position and depth; other player-state attributes come from player inputs, and no component determines hits/damage/deaths.
Abstract (from arXiv)
Multiplayer world models must generate independently controlled views with consistent representations of both players and their shared environment. Most existing approaches coordinate multiple players through joint multi-view generation, whose cost grows with each additional player. We present WorldCast, a distributed multiplayer world model in which each player runs a local client comprising a video generator and a state model. Using recorded player positions and map geometry during training, the state model estimates the player's position from generated video and control inputs. Clients exchange player states and project them into camera-aligned player state fields that guide where and how other players are rendered. Shared scene state enables clients to reuse one another's generated observations to maintain consistent scene appearance across views. Experiments on Counter-Strike 2 demonstrate WorldCast's consistency, real-time performance, and distributed scalability. The camera-aligned player state field improves player rendering rates by over an order of magnitude over joint-generation methods, while shared scene state improves visual consistency over whole rounds. Each client runs in real time and exchanges only player and scene states, enabling scalable multiplayer generation without a centralized computational bottleneck. Image quality remains stable over hour-long rollouts.
Related papers
- PAWA: Panoramic Action-conditioned World Modeling for Multiple Agents
- Multics: Generative Multi-Player World Model with Hierarchical Communication
- OverSeer: Multi-Agent World Models Beyond What Agents See
- Esports World Model Dataset
- ActionRoPE: action as coordinates for spatially consistent control in planar-camera game world models
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-08 · Indexed by Pub-AI: 2026-10-09 UTC
WorldCast: Distributed Multiplayer World Models
Preprint posted: 2026-10-08 · Original record ↗
- arxiv
- 2610.12412
Explicit source versions (1)
- v1 · 2026-10-08
- Preprint available
Source date: 2026-10-08 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Multiplayer world models must generate independently controlled views with consistent representations of both players and their shared environment. Most existing approaches coordinate multiple players through joint multi-view generation, whose cost grows with each additional player. We present WorldCast, a distributed multiplayer world model in which each player runs a local client comprising a video generator and a state model. Using recorded player positions and map geometry during training, the state model estimates the player's position from generated video and control inputs. Clients exchange player states and project them into camera-aligned player state fields that guide where and how other players are rendered. Shared scene state enables clients to reuse one another's generated observations to maintain consistent scene appearance across views. Experiments on Counter-Strike 2 demonstrate WorldCast's consistency, real-time performance, and distributed scalability. The camera-aligned player state field improves player rendering rates by over an order of magnitude over joint-generation methods, while shared scene state improves visual consistency over whole rounds. Each client runs in real time and exchanges only player and scene states, enabling scalable multiplayer generation without a centralized computational bottleneck. Image quality remains stable over hour-long rollouts.