MagCache: Fast Video Generation with Magnitude-Aware Cache

TL;DR

MagCache introduces a magnitude-aware caching strategy that adaptively skips unimportant timesteps in video diffusion models, significantly accelerating generation while maintaining high visual quality and requiring minimal calibration.

Contribution

The paper presents a novel magnitude law across models and prompts, enabling a robust, adaptive caching method that outperforms existing acceleration techniques with minimal calibration.

Findings

Achieves 2.10x-2.68x speedup on multiple video models.

Maintains superior visual fidelity compared to existing methods.

Requires only a single sample for calibration.

Abstract

Existing acceleration techniques for video diffusion models often rely on uniform heuristics or time-embedding variants to skip timesteps and reuse cached features. These approaches typically require extensive calibration with curated prompts and risk inconsistent outputs due to prompt-specific overfitting. In this paper, we introduce a novel and robust discovery: a unified magnitude law observed across different models and prompts. Specifically, the magnitude ratio of successive residual outputs decreases monotonically, steadily in most timesteps while rapidly in the last several steps. Leveraging this insight, we introduce a Magnitude-aware Cache (MagCache) that adaptively skips unimportant timesteps using an error modeling mechanism and adaptive caching strategy. Unlike existing methods requiring dozens of curated samples for calibration, MagCache only requires a single sample for calibration. Experimental results show that MagCache achieves 2.10x-2.68x speedups on Open-Sora, CogVideoX, Wan 2.1, and HunyuanVideo, while preserving superior visual fidelity. It significantly outperforms existing methods in LPIPS, SSIM, and PSNR, under similar computational budgets.