MixCache: Mixture-of-Cache for Video Diffusion Transformer Acceleration

2025-08-18

AI Summary

Existing video diffusion Transformer (DiT) caching methods operate at a single granularity, failing to simultaneously optimize generation quality and inference speed. This work proposes MixCache—a training-free caching framework enabling collaborative multi-granularity optimization. Its core contributions are threefold: (1) a context-aware cache triggering mechanism; (2) an adaptive hybrid decision strategy that dynamically selects the optimal caching granularity—per-step, per-classifier-free-guidance (CFG) scale, or per-module—based on redundancy analysis; and (3) the first joint scheduling scheme for multi-granularity caches. Evaluated on Wan 14B and HunyuanVideo, MixCache achieves 1.94× and 1.97× inference speedup, respectively, while preserving or even improving generation fidelity. By decoupling granularity selection from fixed architectural assumptions and enabling fine-grained reuse, MixCache significantly enhances both the efficiency and practical deployability of video DiT models.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Mixture of Experts (MoE)
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Leveraging the Transformer architecture and the diffusion process, video DiT models have emerged as a dominant approach for high-quality video generation. However, their multi-step iterative denoising process incurs high computational cost and inference latency. Caching, a widely adopted optimization method in DiT models, leverages the redundancy in the diffusion process to skip computations in different granularities (e.g., step, cfg, block). Nevertheless, existing caching methods are limited to single-granularity strategies, struggling to balance generation quality and inference speed in a flexible manner. In this work, we propose MixCache, a training-free caching-based framework for efficient video DiT inference. It first distinguishes the interference and boundary between different caching strategies, and then introduces a context-aware cache triggering strategy to determine when caching should be enabled, along with an adaptive hybrid cache decision strategy for dynamically selecting the optimal caching granularity. Extensive experiments on diverse models demonstrate that, MixCache can significantly accelerate video generation (e.g., 1.94$ imes$ speedup on Wan 14B, 1.97$ imes$ speedup on HunyuanVideo) while delivering both superior generation quality and inference efficiency compared to baseline methods.

Problem

• High computational cost in video DiT models
• Limited single-granularity caching strategies
• Balancing generation quality and inference speed

Innovation

• Training-free caching framework for video DiT
• Context-aware cache triggering strategy
• Adaptive hybrid cache decision strategy