ShareVerse: Multi-Agent Consistent Video Generation for Shared World Modeling

arXiv 26.3 2026 48.3 method, benchmark, application

TLDR

ShareVerse enables multi-agent consistent video generation for shared world modeling using spatial concatenation and cross-agent attention on CARLA data.

Reasoning

The paper presents a novel framework for multi-agent consistent video generation, with strengths in spatial concatenation and cross-agent attention for shared world consistency. However, it relies solely on simulated CARLA data and lacks real-world validation, limiting its generalizability.

Read-first score

Read-first score 48.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 48.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Topical relevance 29%

68.6

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=dataset,metric

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=dataset

Citation impact 18%

28.1

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.28055904

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 331.

Keyword Scores

world model

video world model

generative world model

interactive world model

world dynamics prediction

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

A dataset for large-scale multi-agent interactive world modeling built on the CARLA simulation platform, featuring diverse scenes, weather conditions, interactive trajectories, and paired multi-view videos (front/rear/left/right views per agent) with camera data.
A spatial concatenation strategy for four-view videos of independent agents to model a broader environment and ensure internal multi-view geometric consistency.
Integration of cross-agent attention blocks into a pretrained video model, enabling interactive transmission of spatial-temporal information across agents for shared world consistency in overlapping regions and reasonable generation in non-overlapping regions.

Methodology

ShareVerse builds a large-scale multi-agent interactive world modeling dataset using the CARLA simulator, with diverse scenes, weather conditions, and interactive trajectories paired with multi-view videos (four views per agent). It proposes a spatial concatenation strategy for four-view videos to model a broader environment and ensure geometric consistency, and integrates cross-agent attention blocks into a pretrained video model to enable interactive spatial-temporal information transmission across agents. The model supports 49-frame large-scale video generation.

Key Results

ShareVerse supports 49-frame large-scale video generation, accurately perceives the position of dynamic agents, and achieves consistent shared world modeling.

Resources

Paper