Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Survey on Cache Methods in Diffusion Models: Toward Efficient Multi-Modal Generation

Jiacheng Liu, Xinyu Wang, Yuqi Lin, Zhikai Wang, Peiru Wang, Peiliang Cai, Qinming Zhou, Zhengan Yan, Zexuan Yan, Zhengyi Shi, Chang Zou, Yue Ma, Linfeng Zhang

TL;DR

The surveyed work presents Diffusion Caching as a training-free, architecture-agnostic approach to reduce inference latency in diffusion models by reusing temporally correlated intermediate computations. It establishes a unified taxonomy dividing static and dynamic caching, with dynamic variants including timestep-, layer-, predictive-, and hybrid strategies, and discusses how these can be integrated with existing acceleration techniques. The paper synthesizes theory and practice, detailing U-Net and DiT-specific caching opportunities, and surveys a broad set of applications across images, video, 3D, audio, and science. It also highlights salient challenges—memory overhead, potential quality degradation, and limited theoretical grounding—and outlines future directions toward principled frameworks and multi-technology integration for scalable, efficient generative AI.

Abstract

Diffusion Models have become a cornerstone of modern generative AI for their exceptional generation quality and controllability. However, their inherent \textit{multi-step iterations} and \textit{complex backbone networks} lead to prohibitive computational overhead and generation latency, forming a major bottleneck for real-time applications. Although existing acceleration techniques have made progress, they still face challenges such as limited applicability, high training costs, or quality degradation. Against this backdrop, \textbf{Diffusion Caching} offers a promising training-free, architecture-agnostic, and efficient inference paradigm. Its core mechanism identifies and reuses intrinsic computational redundancies in the diffusion process. By enabling feature-level cross-step reuse and inter-layer scheduling, it reduces computation without modifying model parameters. This paper systematically reviews the theoretical foundations and evolution of Diffusion Caching and proposes a unified framework for its classification and analysis. Through comparative analysis of representative methods, we show that Diffusion Caching evolves from \textit{static reuse} to \textit{dynamic prediction}. This trend enhances caching flexibility across diverse tasks and enables integration with other acceleration techniques such as sampling optimization and model distillation, paving the way for a unified, efficient inference framework for future multimodal and interactive applications. We argue that this paradigm will become a key enabler of real-time and efficient generative AI, injecting new vitality into both theory and practice of \textit{Efficient Generative Intelligence}.

A Survey on Cache Methods in Diffusion Models: Toward Efficient Multi-Modal Generation

TL;DR

The surveyed work presents Diffusion Caching as a training-free, architecture-agnostic approach to reduce inference latency in diffusion models by reusing temporally correlated intermediate computations. It establishes a unified taxonomy dividing static and dynamic caching, with dynamic variants including timestep-, layer-, predictive-, and hybrid strategies, and discusses how these can be integrated with existing acceleration techniques. The paper synthesizes theory and practice, detailing U-Net and DiT-specific caching opportunities, and surveys a broad set of applications across images, video, 3D, audio, and science. It also highlights salient challenges—memory overhead, potential quality degradation, and limited theoretical grounding—and outlines future directions toward principled frameworks and multi-technology integration for scalable, efficient generative AI.

Abstract

Diffusion Models have become a cornerstone of modern generative AI for their exceptional generation quality and controllability. However, their inherent \textit{multi-step iterations} and \textit{complex backbone networks} lead to prohibitive computational overhead and generation latency, forming a major bottleneck for real-time applications. Although existing acceleration techniques have made progress, they still face challenges such as limited applicability, high training costs, or quality degradation. Against this backdrop, \textbf{Diffusion Caching} offers a promising training-free, architecture-agnostic, and efficient inference paradigm. Its core mechanism identifies and reuses intrinsic computational redundancies in the diffusion process. By enabling feature-level cross-step reuse and inter-layer scheduling, it reduces computation without modifying model parameters. This paper systematically reviews the theoretical foundations and evolution of Diffusion Caching and proposes a unified framework for its classification and analysis. Through comparative analysis of representative methods, we show that Diffusion Caching evolves from \textit{static reuse} to \textit{dynamic prediction}. This trend enhances caching flexibility across diverse tasks and enables integration with other acceleration techniques such as sampling optimization and model distillation, paving the way for a unified, efficient inference framework for future multimodal and interactive applications. We argue that this paradigm will become a key enabler of real-time and efficient generative AI, injecting new vitality into both theory and practice of \textit{Efficient Generative Intelligence}.
Paper Structure (37 sections, 63 equations, 2 figures, 1 table)

This paper contains 37 sections, 63 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background
  3. Challenges
  4. Motivation
  5. Contributions
  6. Related Methods
  7. Distillation
  8. Pruning & Sparsification
  9. Parallel Sampling & Hardware Acceleration
  10. Sampler Optimization
  11. Cache Acceleration
  12. Taxonomy of Cache Acceleration
  13. Diffusion Theory
  14. Caching Acceleration Techniques
  15. U-Net Architecture
  16. ...and 22 more sections

Figures (2)

  • Figure 1: The development trend of Diffusion Caching The line chart illustrates the rapid growth in the number of related works from 2024 to the present. Representative works from each period are highlighted.
  • Figure 2: Overview of Diffusion Cache Framework The upper panel illustrates our unified taxonomy that categorizes caching methods into Static and Dynamic paradigms, with Dynamic Caching further divided into Timestep-Adaptive, Layer-Adaptive, Predictive, and Hybrid strategies. The lower panel shows applications across diverse generative tasks.