Less is Enough: Training-Free Video Diffusion Acceleration via Runtime-Adaptive Caching

2025-07-03

AI Summary

Video generation diffusion models suffer from slow inference and high computational overhead due to iterative denoising, severely hindering practical deployment. To address this, we propose EasyCache—a training-free, architecture-agnostic runtime adaptive caching framework. Its core innovation lies in a lightweight online feature vector caching and dynamic reuse mechanism that automatically identifies and exploits redundant computations solely via real-time statistics, eliminating the need for offline analysis or hyperparameter tuning. EasyCache is compatible with mainstream architectures including OpenSora, Wan2.1, and HunyuanVideo. Under identical generation quality, it achieves 2.1–3.3× inference speedup and up to 36% PSNR improvement, significantly outperforming existing acceleration methods.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Hardware-aware ML
• Natural Language Processing: Generation

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Video generation models have demonstrated remarkable performance, yet their broader adoption remains constrained by slow inference speeds and substantial computational costs, primarily due to the iterative nature of the denoising process. Addressing this bottleneck is essential for democratizing advanced video synthesis technologies and enabling their integration into real-world applications. This work proposes EasyCache, a training-free acceleration framework for video diffusion models. EasyCache introduces a lightweight, runtime-adaptive caching mechanism that dynamically reuses previously computed transformation vectors, avoiding redundant computations during inference. Unlike prior approaches, EasyCache requires no offline profiling, pre-computation, or extensive parameter tuning. We conduct comprehensive studies on various large-scale video generation models, including OpenSora, Wan2.1, and HunyuanVideo. Our method achieves leading acceleration performance, reducing inference time by up to 2.1-3.3$ imes$ compared to the original baselines while maintaining high visual fidelity with a significant up to 36% PSNR improvement compared to the previous SOTA method. This improvement makes our EasyCache a efficient and highly accessible solution for high-quality video generation in both research and practical applications. The code is available at https://github.com/H-EmbodVis/EasyCache.

Problem

• Accelerate video diffusion models without training
• Reduce redundant computations in denoising process
• Maintain high visual fidelity during inference

Innovation

• Training-free video diffusion acceleration framework
• Runtime-adaptive caching mechanism
• Reuses computed transformation vectors dynamically