HY-World 2.0: A Multi-Modal World Model for Reconstructing, Generating, and Simulating 3D Worlds

arXiv 2026 43.9 method

TLDR

HY-World 2.0 is a multi-modal world model that generates, reconstructs, and simulates 3D worlds from text, images, and videos, achieving state-of-the-art performance on benchmarks.

Reasoning

The paper presents a comprehensive framework with clear methodological stages and innovations, and it validates performance on benchmarks, indicating strong empirical grounding. However, the abstract lacks details on limitations and specific benchmark results, and the connection to world dynamics prediction or RL is not evident.

Read-first score

Read-first score 43.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Recency 6%

100

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

Methodology quality 18%

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

Topical relevance 29%

64.3

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=code

Citation impact 18%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 12%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 350.

Keyword Scores

world model

world simulator

generative world model

interactive world model

video world model

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

HY-Pano 2.0 for enhanced panorama fidelity in world generation
WorldNav for trajectory planning enabling 3D scene understanding and navigation
WorldStereo 2.0 with consistent memory for keyframe-based view generation
WorldMirror 2.0 with refined architecture and learning strategy for 3D reconstruction from multi-view images or videos
WorldLens, a high-performance 3DGS rendering platform with engine-agnostic architecture, automatic IBL lighting, collision detection, and training-rendering co-design

Methodology

HY-World 2.0 is a multi-modal world model that accepts text, single-view images, multi-view images, or videos, and outputs 3D Gaussian Splatting scenes. For generative inputs (text or single image), it uses a four-stage pipeline: panorama generation (HY-Pano 2.0), trajectory planning (WorldNav), world expansion (WorldStereo 2.0), and world composition (WorldMirror 2.0). WorldLens provides interactive rendering with character support.

Key Results

HY-World 2.0 achieves state-of-the-art performance on multiple benchmarks among open-source approaches and delivers results comparable to the closed-source model Marble.

Resources

Paper