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

2026-05-29

AI Summary

This work addresses the inference efficiency bottlenecks in interactive video world models for long-horizon generation, which stem from context inflation, high computational costs of attention, and redundant denoising steps. The authors propose a training-free, adaptive inference acceleration framework that uniquely integrates user interaction signals into computational scheduling to dynamically manage spatial memory, adjust temporal context length, and reuse historical denoising outputs. By combining adaptive context pruning, a denoising cache mechanism, and hardware-aware 3D block-sparse attention implemented via Triton-fused kernels, the method achieves up to 2.59× speedup on HY-WorldPlay and Matrix-Game-3.0 while preserving high visual generation quality.

Abstract

Interactive video world models generate video chunk by chunk in response to user-controlled camera movements, enabling applications such as real-time game simulation, virtual scene navigation, and embodied AI training. However, scaling to long interactive trajectories is prohibitively expensive due to growing context memory, quadratic attention complexity, and repeated denoising steps. We present Light Interaction, a training-free inference acceleration framework for interactive video world models. Our key insight is that interaction naturally enables trajectory-dependent adaptive computation: retrieved spatial memory can be discarded during novel exploration, temporal context can be adjusted according to local latent dynamics, and early-step model outputs can be reused when the camera revisits familiar regions. Based on this insight, Light Interaction combines adaptive context management, denoising cache acceleration, and hardware-software co-designed 3D block sparse attention with fused Triton kernels. Evaluated on HY-WorldPlay and Matrix-Game-3.0, Light Interaction achieves up to 2.59x speedup without model retraining while maintaining competitive visual quality.

Problem

• interactive video world models
• inference acceleration
• long interactive trajectories
• context memory
• quadratic attention complexity

Innovation

• training-free acceleration
• interactive video world models
• adaptive context management
• denoising cache
• 3D block sparse attention