MoVerse: Real-Time Video World Modeling with Panoramic Gaussian Scaffold

arXiv 2026 54.5 method, system

TLDR

MoVerse creates a real-time interactive 360° video world model from a single image using panoramic Gaussian scaffold and distilled generative video renderer.

Reasoning

Strengths include a novel combination of explicit 3D representation and generative video for real-time interactive world creation from a single image. Weaknesses are the lack of explicit comparisons to baselines and potential limitations in complex scenes, though the abstract demonstrates practical performance.

Read-first score

Read-first score 54.5, 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%

91.2

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.91200371

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=none

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Citation velocity 12%

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

Field roles

FoundationFrontierBridge

Rank sensitivity

Stability: volatile; rank range: 491.

Keyword Scores

video world model

world model

interactive world model

generative world model

world simulator

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Separating world construction from observation rendering to enable interactive navigation from a single narrow-FOV image.
Expanding the input into a gravity-aligned 360° panorama using topology-aware diffusion to complete the missing field of view before 3D reasoning.
Lifting the panorama into a persistent 3D Gaussian scaffold via panoramic geometry-aware residual prediction for dense, renderable spatial memory.
Gaussian-conditioned video renderer trained by distilling a bidirectional diffusion teacher into a causal autoregressive student for bounded-latency streaming.

Methodology

MoVerse first expands a single narrow-FOV image into a gravity-aligned 360° panorama using topology-aware diffusion. It then lifts the panorama into a persistent 3D Gaussian scaffold via panoramic geometry-aware residual prediction. Finally, a Gaussian-conditioned video renderer, trained by distilling a bidirectional diffusion teacher into a causal autoregressive student, generates photorealistic video along user-specified camera trajectories.

Key Results

MoVerse supports real-time scene roaming at 8 FPS on a single NVIDIA RTX 4090 GPU, demonstrating a practical path toward single-image world creation with interactive video output.

Resources

Paper