Precise Editing and Flexible Referencing for Interactable Worlds

arXiv 2026 39.6 method

TLDR

EditWorld is a video world model enabling precise editing and flexible referencing in interactable worlds, with novel attention mechanisms and a benchmark for streaming world editing.

Reasoning

The paper clearly addresses a gap in world modeling by extending it from exploration to precise modification and introduces targeted mechanisms such as Gated Causal Attention and Sparse Context. Weaknesses include the abstract lacking discussion of limitations, generalization, or broader comparisons, with evidence primarily limited to the proposed benchmark.

Read-first score

Read-first score 39.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 49.

Recency 5%

100

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

Topical relevance 25%

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=benchmark,evaluation,metric

Reproducibility 15%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=github

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

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 465.

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

Extends video world models from exploration to precise modification by streaming editing instructions and reference images during autoregressive generation.
Introduces Gated Causal Attention for temporally varying editing conditions and reference images.
Introduces a Sparse Context mechanism that maintains a bounded historical context for long-horizon inference.
Adopts joint autoregressive and bidirectional training with annealed self-resampling.
Constructs a dedicated data synthesis and annotation pipeline providing supervision for world editing.
Presents WBench-Editing to systematically evaluate streaming world editing capabilities.

Methodology

EditWorld is a video world model that streams editing instructions and reference images as conditions during autoregressive generation. It uses Gated Causal Attention to handle temporally varying editing conditions and reference images, plus a Sparse Context mechanism to bound historical context during long-horizon inference. Training uses joint autoregressive and bidirectional objectives with annealed self-resampling, supported by a custom data synthesis and annotation pipeline, and evaluation is conducted on the proposed WBench-Editing benchmark.

Key Results

EditWorld achieves the best overall performance on WBench-Editing with an overall score of 73.8 and an editing score of 80.0, substantially outperforming existing methods on editing-related metrics.

Resources

Paper