FantasyWorld: Geometry-Consistent World Modeling via Unified Video and 3D Prediction

arXiv 25.9 2025 59.3 method

TLDR

FantasyWorld augments frozen video models with a geometric branch for joint video and 3D prediction, achieving geometry-consistent world modeling.

Reasoning

The paper introduces a novel framework that combines video and 3D prediction with cross-branch supervision, showing strong empirical results. However, the abstract lacks details on real-world datasets and does not address interactive or RL-specific applications.

Read-first score

Read-first score 59.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 37.

Methodology quality 25%

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

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Topical relevance 42%

52.9

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 25%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 279.

Keyword Scores

world model

video world model

generative world model

world dynamics prediction

world simulator

model-based reinforcement learning world model

interactive world model

Deep Analysis

Innovations

Augmenting frozen video foundation models with a trainable geometric branch for joint video and 3D modeling
Cross-branch supervision where geometry cues guide video generation and video priors regularize 3D prediction
Single forward pass joint modeling of video latents and an implicit 3D field
Versatile 3D representations from the geometric branch enabling downstream tasks like novel view synthesis and navigation without per-scene optimization

Methodology

FantasyWorld augments frozen video foundation models with a trainable geometric branch, enabling joint modeling of video latents and an implicit 3D field in a single forward pass. Cross-branch supervision is used where geometry cues guide video generation and video priors regularize 3D prediction, yielding consistent and generalizable 3D-aware video representations.

Key Results

FantasyWorld outperforms recent geometry-consistent baselines in multi-view coherence and style consistency. Ablation studies confirm that these gains stem from the unified backbone and cross-branch information exchange.

Resources

Paper