Multi-player interactive world models aim to simulate shared environments from multiple players' perspectives, but achieving robust generation while scaling to more players remains challenging. Existing approaches that rely on fully-connected communication (e.g., Solaris) incur quadratic attention costs as the number of players grows, while centralized approaches (e.g., γ-World) suffer from interference between unrelated events in a shared global representation. To address these challenges, we present Multics, a multi-player video generation framework with hierarchical communication that structures information exchange across player, group, and global levels. This hierarchy introduces group tokens into self-attention as intermediate group-level representations that aggregate locally relevant player information. These tokens connect to global tokens for cross-group communication, while the hierarchical topology limits direct mixing between unrelated local events. Group-token parameters are shared across group slots, allowing the same communication module to accommodate varying numbers of players without introducing new group-specific parameters. We further propose budget-constrained dynamic grouping based on recent camera trajectories, which adaptively groups relevant players while limiting unnecessary communication. Experiments demonstrate that Multics achieves state-of-the-art performance in 4- and 6-player settings, with improved cross-player consistency, player-specific action and camera controllability, and interference robustness. Moreover, Multics generalizes from 4-player training to 6-, 8-, and 10-player inference without retraining, while reducing self-attention computation compared with fully-connected communication.
Figures & tables
Figure 1: Multics is a multi-player video generation model with hierarchical communication that structures information exchange across player, group, and global levels. Multics can generate robust multi-player videos with cross-player consistency, player-specific controllability, and interference robustness, and can be generalized to more players without retraining.
Figure 2: Overview of Multics . Left : The training pipeline of Multics . The video diffusion model is trained to jointly generate multi-player videos V={Vi}i=1N , conditioned on each player i ’s initial image Ii , action sequence Ai , camera trajectory Ci , and text prompt Pi . Right : Comparison of different communication methods. (a) Hierarchical communication in our method; (b) Fully-connected communication in Savva et al. (2026) ; (c) Centralized communication in Liu et al. (2026a) ; (d) Group-based extension of centralized communication with direct intra-group player connections. The matrices illustrate attention connectivity at one latent timestep; block-causal temporal masking and spatial-token expansion are omitted for clarity.
Method
Consistency
Action Accuracy
Camera Accuracy
Visual Quality
GeoKPM ↑
PSNR ↑
SSIM ↑
LPIPS ↓
IDM ↑
RotErr ↓
TransErr ↓
CamMC ↓
FID ↓
FVD ↓
Top: Quantitative Comparison
4 Players
Solaris ( Savva et al., 2026 )
48.51%
17.62
0.4952
0.4324
93.87
2.572
13.50
15.00
28.41
171.9
γ -World ( Liu et al., 2026a )
49.42%
17.76
0.5009
0.4293
93.98
2.728
13.66
15.26
29.36
200.5
Ours
54.17%
18.10
0.5005
0.4184
94.17
2.434
11.67
13.21
26.39
162.3
Table 1: Quantitative comparison with previous methods . We compare Multics with Solaris ( Savva et al., 2026 ) and γ -World ( Liu et al., 2026a ) on consistency, action accuracy, camera accuracy, and visual quality. Top : Results with 4 and 6 players. Middle : Generalization to more players without additional training, where 4→N denotes inference with N players using models trained with 4 players. Bottom : Results on EgoTeam. Bold numbers indicate the best result within each setting.
Figure 3: Qualitative comparison of Multics and fully-connected and centralized baselines in 4-/6-player settings. We compare Multics with Solaris ( Savva et al., 2026 ) and γ -World ( Liu et al., 2026a ) on (a) cross-player consistency: players observing overlapping views of the same scene and one another; (b) player-specific controllability: adherence to aim-down-sights (ADS) and no-firing controls; and (c) interference robustness: independence of unrelated concurrent events and preservation of a player’s scene upon revisiting despite other players’ movements. Top-down maps are provided for visualization only, and are not used as conditioning inputs. Other players are omitted for clarity, and can be found in supplementary videos.
Figure 4: Qualitative comparison of Multics and fully-connected and centralized baselines in generalizing to 6-/8-/10-player scenarios.
Method
Consistency
Action Accuracy
Camera Accuracy
Visual Quality
GeoKPM ↑
PSNR ↑
SSIM ↑
LPIPS ↓
IDM ↑
RotErr ↓
TransErr ↓
CamMC ↓
FID ↓
FVD ↓
Group-based
48.26%
17.50
0.4878
0.4317
94.02
2.780
13.18
14.91
28.76
158.5
Fixed grouping
50.71%
18.16
0.5052
0.4197
94.05
2.496
12.09
13.71
27.78
168.0
Random grouping
46.76%
17.78
0.5036
0.4329
93.50
4.796
14.61
17.90
29.83
208.2
Ours
54.17%
18.10
0.5005
0.4184
94.17
2.434
11.67
13.21
26.39
162.3
Table 2: Ablation on grouping strategies.
Figure 5: Qualitative ablation study on communication mechanisms. The group-based method exhibits inconsistencies when revisiting previously observed scenes (top) and across player views (bottom).
Figure 6: Comparisons of computational efficiency. Multics reduces computation relative to fully-connected communication and achieves efficiency comparable to centralized and group-based methods.
Figure 7: Effect of communication budget on GeoKPM.
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 8: Generalization from 4 to 6 players. Relative performance degradation of models trained with 4 players compared with the same methods trained with 6 players, all evaluated with 6 players. Lower values indicate better preservation of performance; negative values indicate improvement. Bold labels mark the smallest degradation for each metric. Our method achieves the smallest degradation on six of the eight reported metrics.
Method
Consistency
Action Accuracy
Camera Accuracy
Visual Quality
GeoKPM ↑
PSNR ↑
SSIM ↑
LPIPS ↓
IDM ↑
RotErr ↓
TransErr ↓
CamMC ↓
FID ↓
FVD ↓
Copy group 0
45.01%
16.43
0.4563
0.4797
94.08
4.413
12.91
16.01
26.43
199.0
Copy group 1
43.63%
16.41
0.4557
0.4799
94.04
4.606
13.15
16.38
26.39
200.2
Average both
43.96%
16.41
0.4557
0.4802
94.04
4.765
13.32
16.62
26.82
202.5
Ours
54.33%
16.97
0.4659
0.4517
94.31
3.760
11.15
13.74
21.80
133.8
Appendix
Table 3: Ablation on designs of group-token parameters.
Figure 9: Qualitative comparison on EgoTeam. We compare Solaris, γ -World, and Multics on multi-robot egocentric video generation.
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.
World models for interactive video generation have largely focused on single-agent settings, where future observations are generated from a single control signal. However, many generated environments require multi-agent interaction: multiple players, robots, or embodied agents act simultaneously within a shared space. Scaling world models to such settings requires a principled multi-agent design: agents should remain independently controllable, permutation-symmetric, and support efficient inference while maintaining consistency across time and perspectives. In this paper, we present our generative multi-agent world model for interactive simulation. It introduces Simplex Rotary Agent Encoding, a parameter-free extension of 3D RoPE that represents agents as vertices of a regular simplex in rotary angle space. This gives each agent a distinct phase while making all agents permutation-equivalent, enabling scalable agent identity without learned per-slot identities or a fixed agent ordering. To avoid dense all-to-all attention across agents, we further propose Sparse Hub Attention, where learnable hub tokens mediate token interaction across agents, reducing cross-agent attention cost from quadratic to linear in the number of agents. For real-time rollout, we distill a full-context diffusion teacher into a causal student that generates temporal blocks sequentially with KV caching, enabling action-responsive generation at 24 FPS. Experiments in multiplayer virtual environments show that our model improves video fidelity, action controllability, and inter-agent consistency over slot-based and dense-attention baselines, while generalizing from two to four players without additional training.
Fangfu Liu, Kai He, Tianchang Shen +7
NVIDIA · Tsinghua University · University of Toronto +1
Existing action-conditioned video generation models (video world models) are limited to single-agent perspectives, failing to capture the multi-agent interactions of real-world environments. We introduce Solaris, a multiplayer video world model that simulates consistent multi-view observations. To enable this, we develop a multiplayer data system designed for robust, continuous, and automated data collection on video games such as Minecraft. Unlike prior platforms built for single-player settings, our system supports coordinated multi-agent interaction and synchronized videos + actions capture. Using this system, we collect 12.64 million multiplayer frames and propose an evaluation framework for multiplayer movement, memory, grounding, building, and view consistency. We train Solaris using a staged pipeline that progressively transitions from single-player to multiplayer modeling, combining bidirectional, causal, and Self Forcing training. In the final stage, we introduce Checkpointed Self Forcing, a memory-efficient Self Forcing variant that enables a longer-horizon teacher. Results show our architecture and training design outperform existing baselines. Through open-sourcing our system and models, we hope to lay the groundwork for a new generation of multi-agent world models.