Prisma-World: Camera-Controllable Multi-Agent Video World Model

arXiv 2026 61.9 method

TLDR

Prisma-World is a camera-controllable multi-agent video world model that ensures cross-view consistency via geometry-aware denoising and a new synthetic dataset.

Reasoning

The paper introduces a novel approach for multi-agent video generation with cross-view consistency, leveraging geometry-aware attention and a curriculum training strategy. Its main weakness is reliance on a synthetic dataset (UE5) without real-world validation, and it does not address interactive agent actions beyond camera control.

Read-first score

Read-first score 61.9, 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

Citation impact 18%

95.2

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.95211968

Methodology quality 18%

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

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

Reproducibility 18%

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

Citation velocity 12%

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

Field roles

FoundationFrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 451.

Keyword Scores

world model

video world model

generative world model

world simulator

world dynamics prediction

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Formulates multi-agent generation as a joint geometry-aware denoising process for cross-view consistency
Processes all agent videos within one full-attention sequence
Multi-agent RoPE design to distinguish agent identities while preserving synchronized temporal coordinates
Injects relative camera geometry into attention to bias overlapping viewpoints toward shared scene evidence
Overlap-decaying curriculum training paradigm
Minimap-conditioned structural guidance
PrismaDataset: large-scale UE5 dataset with panoramic acquisition, composable multi-agent view groups, and precise camera/action annotations

Methodology

Prisma-World processes all agent videos within one full-attention sequence, uses a multi-agent RoPE design to distinguish agent identities while preserving synchronized temporal coordinates, and injects relative camera geometry into attention to bias overlapping viewpoints toward shared scene evidence. It also employs an overlap-decaying curriculum training paradigm and minimap-conditioned structural guidance to strengthen multi-view consistency and global spatial perception.

Key Results

A single Prisma-World model generates high-fidelity multi-agent videos with flexible agent numbers, camera controllability, improved cross-view consistency, and spatial grounding under minimap guidance.

Resources

Paper