Lyra 2.0: Explorable Generative 3D Worlds

arXiv 2026 31.1 method

TLDR

Lyra 2.0 generates persistent, explorable 3D worlds using video generation and 3D reconstruction, addressing spatial forgetting and temporal drifting.

Reasoning

The paper proposes a novel framework for long-horizon 3D scene generation from video, with clear problem identification and solutions. However, the abstract is incomplete and lacks explicit empirical validation or real-world benchmarks.

Read-first score

Read-first score 31.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 33.

Recency 5%

100

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

Methodology quality 15%

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

Topical relevance 25%

47.1

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 15%

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

Graph prestige 15%

5.7

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0016

Citation impact 15%

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

Citation velocity 10%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 224.

Keyword Scores

video world model

generative world model

world model

world simulator

interactive world model

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Per-frame 3D geometry used solely for information routing — retrieving relevant past frames and establishing dense correspondences with target viewpoints — to combat spatial forgetting while relying on the generative prior for appearance synthesis.
Self-augmented history training that exposes the model to its own degraded outputs to teach it to correct temporal drifting rather than propagate it.
Integration of these techniques to generate substantially longer, 3D-consistent camera trajectory videos, which are then used to fine-tune feed-forward reconstruction models for high-quality 3D scene recovery.

Methodology

Lyra 2.0 generates camera-controlled videos simulating scene walkthroughs and lifts them to 3D via feed-forward reconstruction. To handle long trajectories with large viewpoint changes and revisits, it maintains per-frame 3D geometry for information routing and trains with self-augmented histories to correct drift. The resulting long, consistent videos are used to fine-tune feed-forward reconstruction models.

Key Results

The method enables substantially longer and 3D-consistent video trajectories, which allow fine-tuned feed-forward reconstruction models to reliably recover high-quality 3D scenes.

Limitations

No explicit limitations of the proposed Lyra 2.0 framework are discussed in the abstract.

Resources

Paper