World in World: Explore the World with World Models

arXiv 2026 38.4 method

TLDR

Training-free inference-time interface for frozen autoregressive video world models enabling controllable camera viewpoints, long-horizon revisiting, and human motion from source videos.

Reasoning

The paper presents a novel unified interface leveraging native self-attention and evidence-aware CFG for controllable video generation, addressing synchronization, completion, and revisiting. However, the abstract lacks visible empirical evaluations or benchmark results, making it difficult to assess demonstrated effectiveness.

Read-first score

Read-first score 38.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 51.

Recency 5%

100

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

Topical relevance 25%

72.9

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

Methodology quality 15%

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

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%

10.9

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0031

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: 414.

Keyword Scores

world model

video world model

generative world model

interactive world model

world simulator

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Introduces World in World, a training-free inference-time interface that adds flexible control to a frozen causal video world model without task-specific modules or additional training.
Converts heterogeneous control evidence into camera- and time-labelled clean visual states that are read through the native self-attention of a frozen causal video model.
Uses multiple evidence sources with token-level support and availability schedules: source-video observations, target-view scene projections, geometry renderings for newly exposed regions, and retrieved generated states beyond the rolling cache.
Proposes a correspondence router that combines persistent point identities with geometry to establish token correspondences, steering supported queries toward matching source-video tokens.
Introduces evidence-wise attention classifier-free guidance (EWA) that independently regulates each auxiliary channel's contribution using attention responses from the same denoising forward pass.
Provides a shared control interface supporting camera-controlled rerendering, long-horizon revisiting, and human-motion transfer with the same frozen backbone.

Methodology

The method is a training-free inference-time interface over a frozen autoregressive causal video model. It converts heterogeneous control evidence into camera- and time-labelled visual states read via native self-attention, supported by a correspondence router linking persistent point identities with geometry and an evidence-wise attention CFG mechanism that regulates auxiliary evidence contributions. Evaluation focuses on camera-controlled video rerendering under diverse viewpoint changes.

Key Results

The paper reports evaluation on camera-controlled video rerendering under diverse viewpoint changes, assessing perceptual quality, temporal consistency, and camera-following accuracy. The abstract does not report quantitative metric values.

Resources

Paper