Light Interaction: Training-Free Inference Acceleration for Interactive Video World Models

arXiv 2026 59.1 method

TLDR

A training-free inference acceleration framework for interactive video world models using adaptive context management and sparse attention.

Reasoning

The paper presents a practical solution to a key bottleneck in interactive video world models, with clear methodology and empirical speedup. However, evaluation is limited to two synthetic benchmarks, and real-world applicability is not demonstrated.

Read-first score

Read-first score 59.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 55.

Recency 6%

100

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

Citation impact 18%

82.7

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.82660481

Topical relevance 29%

78.6

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 18%

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

Reproducibility 18%

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

Citation velocity 12%

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

Field roles

FoundationFrontierBridge

Rank sensitivity

Stability: volatile; rank range: 485.

Keyword Scores

interactive world model

video world model

world model

generative world model

world simulator

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Adaptive context management that discards spatial memory during novel exploration and adjusts temporal context based on local latent dynamics
Denoising cache acceleration that reuses early-step model outputs when the camera revisits familiar regions
Hardware-software co-designed 3D block sparse attention with fused Triton kernels

Methodology

Light Interaction is a training-free inference acceleration framework for interactive video world models. It combines adaptive context management, denoising cache acceleration, and 3D block sparse attention with fused Triton kernels. The framework is evaluated on HY-WorldPlay and Matrix-Game-3.0 datasets, measuring speedup and visual quality.

Key Results

Light Interaction achieves up to 2.59x speedup without model retraining while maintaining competitive visual quality on HY-WorldPlay and Matrix-Game-3.0.

Resources

Paper