๐ŸŽจ Awesome Diffusion Acceleration Cache ๐Ÿš€

July 23, 2026 ยท View on GitHub

๐ŸŽจ Awesome Diffusion Acceleration Cache ๐Ÿš€

Awesome Papers: 147 Verified: 2026-07-23 Metadata: arXiv synced

A verified and taxonomy-driven collection of cache, reuse, and forecasting methods for efficient generative diffusion and flow-matching inference.

Titles, author order, submission dates, revision dates, categories, DOI, and journal references are synchronized from arXiv.

๐Ÿงญ Quick Navigation

โœจ At a Glance

Total papers: 147 2025โ€“2026: 121 Revised papers: 73 Core methods: 99 Systems: 6 Adjacent: 39

Image: 70 Video: 37 3D / 4D: 5 World models: 9 WAM: 2 Diffusion LLM: 19 Policy: 6

๐Ÿ”Ž Curation and Date Policy

  • Core scope: skip, reuse, forecast, compress, or share intermediate computation during iterative diffusion or flow-matching inference.
  • Adjacent scope: image/video/3D generation, world models, World Action Models (WAMs), diffusion policies, scientific diffusion, and diffusion language models are labeled explicitly.
  • Authoritative metadata: title, complete author order, first submission, latest revision, primary category, DOI, and journal reference come from the arXiv API.
  • Date convention: papers are grouped by their first arXiv submission dateโ€”not the arXiv identifier prefix, announcement month, or an unverified venue date.
  • Anti-hallucination rule: entries without a public primary source are excluded; uncertain taxonomy values remain unspecified instead of being guessed.
  • Canonical data: full author lists and metadata live in Paperlists.csv. Run python scripts/sync_arxiv_metadata.py before regenerating this page.

๐Ÿงฉ Taxonomy

The labels deliberately separate what is cached, how it is reused, and when it is refreshed.

Visual labelAxisExamples
Method: TeaCacheMethod nameTeaCache, ToCa, TaylorSeer
Scope: videoTask or modalityimage, video, 3D, world model, WAM, policy, audio, diffusion LLM
Unit: featureCached objectfeature, block/layer, token, attention/KV, latent
Reuse: forecastingReuse mechanismdirect reuse, delta reuse, forecasting, correction
Policy: adaptiveRefresh policyfixed, similarity-aware, error-aware, search-based
Training: training-freeExtra optimizationtraining-free, learned, system

๐Ÿ“ Papers

๐Ÿ—“ 2026

2026 ยท 57 papers

  • [147] Evolving Cache Schedules for Fast Diffusion Policy Inference

    Siying Wang; Kangye Ji; Di Wang; Fei Cheng

    Paper ยท Code ยท submitted 2026-07-22 ยท revised same as submission ยท cs.CV

    Method: EVO Scope: policy Reuse: direct reuse Policy: budget/search

  • [146] DiTango: Cost-Effective Parallel Diffusion Generation with Selective Attention State Reuse

    Yuyang Chen; Runxin Zhong; Zan Zong; Hengjie Li; Yuyang Jin; Jidong Zhai

    Paper ยท DOI ยท submitted 2026-07-17 ยท revised same as submission ยท cs.CV

    Method: DiTango Scope: image Unit: attention/KV Reuse: direct reuse Policy: system-aware

    Published reference: The 35th International Symposium on High-Performance Parallel and Distributed Computing (HPDC'26), July 13--16, 2026, Cleveland, OH, USA

  • [145] LaCache: Exact Caching and Precision-Adaptive Inference for Diffusion Large Language Models

    Xingru Chen; Zelang Liang; Yongjia Ma; Jiqing Zhan; Shuling Yang; Lian Wen et al.

    Paper ยท submitted 2026-07-16 ยท revised 2026-07-22 ยท cs.AI

    Method: LaCache Scope: diffusion LLM Reuse: direct reuse Policy: adaptive

  • [144] CODA: Algorithm-Hardware Co-design for Edge Video Diffusion via NMP-Enabled Compute-Cache Operator Disaggregation

    Yuanpeng Zhang; YuXuan Wu; Yitong Xiao; Chenhao Xue; Yi Ren; Cong Li et al.

    Paper ยท submitted 2026-07-16 ยท revised same as submission ยท cs.AR

    Method: CODA Scope: video Reuse: direct reuse Policy: system-aware Training: system

  • [143] ACID: Adaptive Caching for vIDeo generation

    Om Agrawal; Saurabh Agarwal; Aditya Akella

    Paper ยท submitted 2026-07-14 ยท revised 2026-07-16 ยท cs.CV

    Method: ACID Scope: video Reuse: direct reuse Policy: adaptive

  • [142] Training-Free Acceleration for Vision-Language-Action Models with Action Caching and Refinement

    Ryuji Oi; Hikari Otsuka; Kosuke Matsushima; Yuki Ichikawa; Masato Motomura; Tatsuya Kaneko et al.

    Paper ยท submitted 2026-07-07 ยท revised same as submission ยท cs.RO

    Method: ActionCache Scope: policy Reuse: direct reuse Training: training-free

  • [141] Omni-Flow: A Unified Workflow Orchestration and Distributed KV Cache Sharing Framework for Multimodal Inference

    Bin Xiao; Jingfu Dong; Changran Wang; Yitian Chen; Xiaoyu Zhao; Yuqi Peng et al.

    Paper ยท submitted 2026-06-30 ยท revised same as submission ยท cs.DC

    Method: Omni-Flow Scope: multimodal Unit: attention/KV Reuse: direct reuse

  • [140] OTCache: Optimal Transport for Geometry-Aware Caching in Diffusion Models

    Huanlin Gao; Fang Zhao; Qiang Hui; Fuyuan Shi; Shaoan Zhao; Yantao Li et al.

    Paper ยท Code ยท submitted 2026-06-30 ยท revised same as submission ยท cs.LG

    Method: OTCache Scope: image Reuse: direct reuse Policy: similarity/geometry-aware

  • [139] NaviCache: Test-Time Self-Calibration Caching for Video Generation

    Zheqi Lv; Zhibo Zhu; Jinke Wang; Qi Tian; Shengyu Zhang; Zhengyu Chen et al.

    Paper ยท submitted 2026-06-25 ยท revised 2026-07-11 ยท cs.CV

    Method: NaviCache Scope: video Reuse: calibration/correction Reuse: direct reuse Policy: adaptive

  • [138] LearniBridge: Learnable Calibration of Feature Caching for Diffusion Models Acceleration

    Xuyue Huang; Zhe Chen; Wang Shen; Xiao-Ping Zhang

    Paper ยท Code ยท submitted 2026-06-25 ยท revised same as submission ยท cs.CV

    Method: LearniBridge Scope: image Unit: feature Reuse: calibration/correction Reuse: learned approximation Reuse: direct reuse Training: learned

  • [137] Chorus II: Cross-Request Sparsity Reuse for Efficient Image-to-Video Generation

    Hao Liu; Chenghuan Huang; Hao Liu; Xing Cai; Chen Li; Ziyang Ma et al.

    Paper ยท submitted 2026-06-23 ยท revised same as submission ยท cs.CV

    Method: Chorus II Scope: video Unit: request Reuse: direct reuse Policy: system-aware

  • [136] Kamera: Unified Position-Invariant Multimodal KV Cache for Training-Free Reuse

    Bole Ma; Jan Eitzinger; Harald Koestler; Gerhard Wellein

    Paper ยท submitted 2026-06-22 ยท revised same as submission ยท cs.DC

    Method: Kamera Scope: multimodal Unit: attention/KV Reuse: direct reuse Training: training-free

  • [135] HERALD: High-Throughput Block Diffusion LLM Serving via CPU-GPU Cooperative KV Cache Retrieval

    Omin Kwon; Doyeon Kim; Jongseok Park; Seung Yul Lee; Ion Stoica; Jae W. Lee

    Paper ยท submitted 2026-06-19 ยท revised same as submission ยท cs.LG

    Method: HERALD Scope: diffusion LLM Unit: attention/KV Reuse: direct reuse Policy: system-aware Training: system

  • [134] MoECa: Aligning Feature Reuse with Expert Decomposition in Diffusion Transformers

    Maoliang Li; Haojing Chen; Jiayu Chen; Zihao Zheng; Xinhao Sun; Hailong Zou et al.

    Paper ยท submitted 2026-06-14 ยท revised same as submission ยท cs.LG

    Method: MoECa Scope: image Unit: feature Reuse: direct reuse

  • [133] Budget-Constrained Step-Level Diffusion Caching

    Mingkun Lei; Tong Zhao; Liangyu Yuan; Chi Zhang

    Paper ยท Code ยท submitted 2026-06-11 ยท revised same as submission ยท cs.CV

    Method: BCS Scope: image Unit: step Reuse: direct reuse Policy: budget/search

  • [132] C3^3ache: Accelerating World Action Models with Cross Inference Chunk Cache

    Weisen Zhao; Lam Nguyen; Zhicong Lu; Yuzhang Shang

    Paper ยท submitted 2026-06-08 ยท revised same as submission ยท cs.LG

    Method: C3ache Scope: WAM Scope: world model Scope: policy Unit: feature Unit: request Reuse: direct reuse Policy: cross-chunk Training: training-free

  • [131] AHA-WAM:Asynchronous Horizon-Adaptive World-Action Modeling with Observation-Guided Context Routing

    Jisong Cai; Long Ling; Shiwei Chu; Zhongshan Liu; Jiayue Kang; Zhixuan Liang et al.

    Paper ยท submitted 2026-06-08 ยท revised same as submission ยท cs.RO

    Method: AHA-WAM Scope: WAM Scope: world model Scope: policy Unit: attention/KV Unit: feature Reuse: direct reuse Policy: hierarchical Policy: rolling-memory Training: learned

  • [130] Less Is More: Training-Free Acceleration Framework of 3D Diffusion Models for Low-Count PET Denoising via Global-Local Trajectory Reduction

    Yuhan Liu; Scott M. Leonard; Marlee Crews; Muhannad Fadhel; Jinkui Hao; Tianqi Chen et al.

    Paper ยท submitted 2026-06-07 ยท revised same as submission ยท cs.CV

    Method: GLTR Scope: 3D Scope: medical Unit: feature Unit: step Reuse: direct reuse Policy: fixed/heuristic Training: training-free

  • [129] ReCache: Learning Budget-Aware Caching Schedules for Diffusion Models via REINFORCE

    Mishan Aliev; Eva Neudachina; Ilya Bykov; Aleksandr Oganov; Kirill Struminsky; Aibek Alanov et al.

    Paper ยท Code ยท submitted 2026-06-04 ยท revised same as submission ยท cs.CV

    Method: ReCache Scope: image Reuse: direct reuse Policy: budget/search Training: learned

  • [128] WaveFilter: Enhancing the Long-Context Capability of Diffusion LLMs via Wavelet-Guided KV Cache Filtering

    Jinnan Yang; Yan Wang; Zhen Bi; Kehao Wu; Xiaojie Li; Jungang Lou et al.

    Paper ยท submitted 2026-05-30 ยท revised same as submission ยท cs.CL

    Method: WaveFilter Scope: diffusion LLM Unit: attention/KV Reuse: direct reuse Policy: frequency-aware

  • [127] Light Interaction: Training-Free Inference Acceleration for Interactive Video World Models

    Jiacheng Lu; Haoyi Zhu; Sipei Yi; Enze Xie; Yu Li; Cheng Zhuo

    Paper ยท submitted 2026-05-29 ยท revised 2026-06-18 ยท cs.CV

    Method: Light Interaction Scope: video Scope: world model Unit: model output Unit: step Reuse: direct reuse Policy: adaptive Policy: system-aware Training: training-free

  • [126] VideoMLA: Low-Rank Latent KV Cache for Minute-Scale Autoregressive Video Diffusion

    Hidir Yesiltepe; Jiazhen Hu; Tuna Han Salih Meral; Adil Kaan Akan; Kaan Oktay; Hoda Eldardiry et al.

    Paper ยท submitted 2026-05-28 ยท revised same as submission ยท cs.CV

    Method: VideoMLA Scope: video Unit: attention/KV Unit: latent Reuse: direct reuse

  • [125] Dynamic-dLLM: Dynamic Cache-Budget and Adaptive Parallel Decoding for Training-Free Acceleration of Diffusion LLM

    Tianyi Wu; Xiaoxi Sun; Yanhua Jiao; Yulin Li; Yixin Chen; YunHao Cao et al.

    Paper ยท Code ยท submitted 2026-05-27 ยท revised same as submission ยท cs.CL

    Method: Dynamic-dLLM Scope: diffusion LLM Reuse: direct reuse Policy: adaptive Policy: budget/search Policy: system-aware Training: training-free

  • [124] Enabling KV Caching of Shared Prefix for Diffusion Language Models

    Younghun Go; Jaehoon Han; Changyong Shin; Chuck Yoo; Gyeongsik Yang

    Paper ยท submitted 2026-05-26 ยท revised 2026-07-02 ยท cs.LG

    Method: โ€” Scope: diffusion LLM Reuse: direct reuse

  • [123] Quantized Keys Steal Attention: Bias Correction for KV-Cache Compression in Video Diffusion

    Tuna Tuncer; Felix Becker; Thomas Pfeil

    Paper ยท submitted 2026-05-25 ยท revised same as submission ยท cs.LG

    Method: โ€” Scope: video Unit: attention/KV Reuse: calibration/correction Reuse: compression Reuse: direct reuse

  • [122] DiSC: Resolution-Scalable Acceleration of Diffusion Models by Exploiting Sparsity and Cached Token Reuse with Hash-based Distribution

    Jieon Yoon; Hangyeol Lee; Jaehoon Heo; Joo-Young Kim

    Paper ยท submitted 2026-05-25 ยท revised same as submission ยท cs.AR

    Method: DiSC Scope: image Unit: token Reuse: direct reuse

  • [121] Trajectory-Consistent Calibration for Cache-Accelerated Diffusion Models

    Mingyu Liang; Dingkun Xu; Jingwei Xu

    Paper ยท submitted 2026-05-24 ยท revised same as submission ยท cs.CV

    Method: TCC Scope: image Reuse: calibration/correction Reuse: direct reuse

  • [120] Accelerating Rectified Flow Models via Trajectory-Aware Caching

    Xiao Liu; Kai Liu; Naiyang Guan; Hongliang Lu; Zhixin Wang; Zhikai Chen et al.

    Paper ยท submitted 2026-05-16 ยท revised same as submission ยท cs.CV

    Method: TACache Scope: image Reuse: calibration/correction Reuse: direct reuse

  • [119] On the Limits of Latent Reuse in Diffusion Models

    Yifeng Yu; Lu Yu

    Paper ยท submitted 2026-05-13 ยท revised same as submission ยท stat.ML

    Method: Latent Reuse Limits Scope: image Unit: latent Reuse: direct reuse

  • [118] Forcing-KV: Hybrid KV Cache Compression for Efficient Autoregressive Video Diffusion Models

    Yicheng Ji; Zhizhou Zhong; Jun Zhang; Qin Yang; XiTai Jin; Ying Qin et al.

    Paper ยท submitted 2026-05-10 ยท revised same as submission ยท cs.CV

    Method: Forcing-KV Scope: video Unit: attention/KV Reuse: compression Reuse: direct reuse

  • [117] FlashClear: Ultra-Fast Image Content Removal via Efficient Step Distillation and Feature Caching

    Yixin Tang; Jiawei Guo; Junxian Li; Zhiteng Li; Jixin Zhao; Bingya Zhang et al.

    Paper ยท submitted 2026-05-09 ยท revised 2026-05-12 ยท cs.CV

    Method: FlashClear Scope: editing Unit: feature Reuse: learned approximation Reuse: direct reuse Training: learned

  • [116] Polestar: Drift-Aware Cache Calibration and Token Commitment for Efficient Inference of Diffusion LLMs

    Mingyu Lee; Akshat Ramachandran; Souvik Kundu; Tushar Krishna

    Paper ยท submitted 2026-05-07 ยท revised same as submission ยท cs.CL

    Method: Polestar Scope: diffusion LLM Unit: token Reuse: calibration/correction Reuse: direct reuse Policy: error/sensitivity-aware

  • [115] DARE: Diffusion Language Model Activation Reuse for Efficient Inference

    Natalia Frumkin; Bokun Wang; Hung-Yueh Chiang; Chi-Chih Chang; Mohamed S. Abdelfattah; Diana Marculescu

    Paper ยท Code ยท submitted 2026-05-01 ยท revised same as submission ยท cs.LG

    Method: DARE Scope: diffusion LLM Unit: feature Reuse: direct reuse

  • [114] Accelerating Frequency Domain Diffusion Models with Error-Feedback Event-Driven Caching

    Dong Liu; Haisheng Wang; Yanxuan Yu

    Paper ยท Code ยท submitted 2026-04-24 ยท revised same as submission ยท cs.LG

    Method: EFCache Scope: image Reuse: calibration/correction Reuse: direct reuse Policy: error/sensitivity-aware Policy: frequency-aware

  • [113] X-Cache: Cross-Chunk Block Caching for Few-Step Autoregressive World Models Inference

    Yixiao Zeng; Jianlei Zheng; Chaoda Zheng; Shijia Chen; Mingdian Liu; Tongping Liu et al.

    Paper ยท submitted 2026-04-22 ยท revised 2026-05-05 ยท cs.CV

    Method: X-Cache Scope: world model Unit: block/layer Reuse: direct reuse

  • [112] Beyond Few-Step Inference: Accelerating Video Diffusion Transformer Model Serving with Inter-Request Caching Reuse

    Hao Liu; Ye Huang; Chenghuan Huang; Zhenyi Zheng; Jiangsu Du; Ziyang Ma et al.

    Paper ยท submitted 2026-04-06 ยท revised same as submission ยท cs.CV

    Method: Inter-Request Cache Scope: video Unit: request Reuse: direct reuse Policy: system-aware Training: system

  • [111] Accelerating Diffusion-based Video Editing via Heterogeneous Caching: Beyond Full Computing at Sampled Denoising Timestep

    Tianyi Liu; Ye Lu; Linfeng Zhang; Chen Cai; Jianjun Gao; Yi Wang et al.

    Paper ยท submitted 2026-03-25 ยท revised same as submission ยท cs.CV

    Method: HeteroCache Scope: video Scope: editing Unit: step Reuse: direct reuse

  • [110] WorldCache: Content-Aware Caching for Accelerated Video World Models

    Umair Nawaz; Ahmed Heakl; Ufaq Khan; Abdelrahman Shaker; Salman Khan; Fahad Shahbaz Khan

    Paper ยท submitted 2026-03-23 ยท revised same as submission ยท cs.CV

    Method: WorldCache Scope: video Scope: world model Reuse: direct reuse

  • [109] EntropyCache: Decoded Token Entropy Guided KV Caching for Diffusion Language Models

    Minsoo Cheong; Donghyun Son; Woosang Lim; Sungjoo Yoo

    Paper ยท Code ยท submitted 2026-03-19 ยท revised same as submission ยท cs.CL

    Method: EntropyCache Scope: diffusion LLM Unit: token Reuse: direct reuse Policy: error/sensitivity-aware

  • [108] WorldCache: Accelerating World Models for Free via Heterogeneous Token Caching

    Weilun Feng; Guoxin Fan; Haotong Qin; Mingqiang Wu; Yuqi Li; Xiangqi Li et al.

    Paper ยท Code ยท submitted 2026-03-06 ยท revised 2026-06-01 ยท cs.CV

    Method: WorldCache Scope: world model Unit: token Reuse: direct reuse

  • [107] Frequency-Aware Error-Bounded Caching for Accelerating Diffusion Transformers

    Guandong Li

    Paper ยท submitted 2026-03-05 ยท revised same as submission ยท cs.CV

    Method: FEBCache Scope: image Reuse: direct reuse Policy: error/sensitivity-aware Policy: frequency-aware

  • [106] SenCache: Accelerating Diffusion Model Inference via Sensitivity-Aware Caching

    Yasaman Haghighi; Alexandre Alahi

    Paper ยท submitted 2026-02-27 ยท revised same as submission ยท cs.CV

    Method: SenCache Scope: image Reuse: direct reuse Policy: error/sensitivity-aware

  • [105] Relational Feature Caching for Accelerating Diffusion Transformers

    Byunggwan Son; Jeimin Jeon; Jeongwoo Choi; Bumsub Ham

    Paper ยท submitted 2026-02-23 ยท revised same as submission ยท cs.CV

    Method: RFC Scope: image Unit: feature Reuse: direct reuse Policy: similarity/geometry-aware

  • [104] SeaCache: Spectral-Evolution-Aware Cache for Accelerating Diffusion Models

    Jiwoo Chung; Sangeek Hyun; MinKyu Lee; Byeongju Han; Geonho Cha; Dongyoon Wee et al.

    Paper ยท submitted 2026-02-22 ยท revised 2026-03-24 ยท cs.CV

    Method: SeaCache Scope: image Reuse: direct reuse Policy: frequency-aware

  • [103] CHAI: CacHe Attention Inference for text2video

    Joel Mathew Cherian; Ashutosh Muralidhara Bharadwaj; Vima Gupta; Anand Padmanabha Iyer

    Paper ยท submitted 2026-02-18 ยท revised same as submission ยท cs.CV

    Method: CHAI Scope: video Reuse: direct reuse

  • [102] AdaCorrection: Adaptive Offset Cache Correction for Accurate Diffusion Transformers

    Dong Liu; Yanxuan Yu; Ben Lengerich; Ying Nian Wu

    Paper ยท submitted 2026-02-13 ยท revised 2026-05-08 ยท cs.CV

    Method: AdaCorrection Scope: image Reuse: calibration/correction Reuse: direct reuse Policy: adaptive

  • [101] Flow caching for autoregressive video generation

    Yuexiao Ma; Xuzhe Zheng; Jing Xu; Xiwei Xu; Feng Ling; Xiawu Zheng et al.

    Paper ยท Code ยท submitted 2026-02-11 ยท revised same as submission ยท cs.CV

    Method: FlowCache Scope: video Reuse: direct reuse

  • [100] FlashBlock: Attention Caching for Efficient Long-Context Block Diffusion

    Zhuokun Chen; Jianfei Cai; Bohan Zhuang

    Paper ยท submitted 2026-02-05 ยท revised 2026-07-03 ยท cs.CV

    Method: FlashBlock Scope: diffusion LLM Unit: attention/KV Reuse: direct reuse

  • [99] DisCa: Accelerating Video Diffusion Transformers with Distillation-Compatible Learnable Feature Caching

    Chang Zou; Changlin Li; Yang Li; Patrol Li; Jianbing Wu; Xiao He et al.

    Paper ยท Code ยท submitted 2026-02-05 ยท revised 2026-04-20 ยท cs.CV

    Method: DisCa Scope: video Unit: feature Reuse: learned approximation Reuse: direct reuse Training: distillation-compatible

  • [98] Quant VideoGen: Auto-Regressive Long Video Generation via 2-Bit KV-Cache Quantization

    Haocheng Xi; Shuo Yang; Yilong Zhao; Muyang Li; Han Cai; Xingyang Li et al.

    Paper ยท Code ยท submitted 2026-02-03 ยท revised 2026-05-05 ยท cs.LG

    Method: Quant VideoGen Scope: video Unit: attention/KV Reuse: compression Reuse: direct reuse

  • [97] Fast Autoregressive Video Diffusion and World Models with Temporal Cache Compression and Sparse Attention

    Dvir Samuel; Issar Tzachor; Matan Levy; Michael Green; Gal Chechik; Rami Ben-Ari

    Paper ยท submitted 2026-02-02 ยท revised 2026-06-12 ยท cs.CV

    Method: โ€” Scope: video Scope: world model Reuse: compression Reuse: direct reuse

  • [96] SPA-Cache: Singular Proxies for Adaptive Caching in Diffusion Language Models

    Wenhao Sun; Rong-Cheng Tu; Yifu Ding; Zhao Jin; Jingyi Liao; Yongcheng Jing et al.

    Paper ยท submitted 2026-01-30 ยท revised 2026-05-23 ยท cs.LG

    Method: SPA-Cache Scope: diffusion LLM Reuse: direct reuse Policy: adaptive

  • [95] Past- and Future-Informed KV Cache Policy with Salience Estimation in Autoregressive Video Diffusion

    Hanmo Chen; Chenghao Xu; Xu Yang; Xuan Chen; Cheng Deng

    Paper ยท submitted 2026-01-29 ยท revised 2026-02-04 ยท cs.CV

    Method: โ€” Scope: video Scope: policy Unit: attention/KV Reuse: direct reuse Policy: error/sensitivity-aware

  • [94] Window-Diffusion: Accelerating Diffusion Language Model Inference with Windowed Token Pruning and Caching

    Fengrui Zuo; Zhiwei Ke; Yiming Liu; Wenqi Lou; Chao Wang; Xuehai Zhou

    Paper ยท Code ยท submitted 2026-01-28 ยท revised 2026-02-01 ยท cs.LG

    Method: Window-Diffusion Scope: diffusion LLM Unit: token Reuse: direct reuse

  • [93] Turbo4DGen: Ultra-Fast Acceleration for 4D Generation

    Yuanbin Man; Ying Huang; Zhile Ren; Miao Yin

    Paper ยท submitted 2026-01-24 ยท revised same as submission ยท cs.GR

    Method: Turbo4DGen Scope: 3D Scope: video Scope: world model Unit: attention/KV Reuse: direct reuse Policy: adaptive

  • [92] Forecast the Principal, Stabilize the Residual: Subspace-Aware Feature Caching for Efficient Diffusion Transformers

    Guantao Chen; Shikang Zheng; Yuqi Lin; Linfeng Zhang

    Paper ยท submitted 2026-01-12 ยท revised same as submission ยท cs.CV

    Method: SubspaceCache Scope: image Unit: feature Reuse: forecasting Reuse: direct reuse

  • [91] RadioDiff-Flux: Efficient Radio Map Construction via Generative Denoise Diffusion Model Trajectory Midpoint Reuse

    Xiucheng Wang; Peilin Zheng; Honggang Jia; Nan Cheng; Ruijin Sun; Conghao Zhou et al.

    Paper ยท submitted 2026-01-06 ยท revised same as submission ยท cs.LG

    Method: RadioDiff-Flux Scope: scientific Unit: step Reuse: direct reuse

๐Ÿ—“ 2025

2025 ยท 64 papers

  • [90] CorGi: Contribution-Guided Block-Wise Interval Caching for Training-Free Acceleration of Diffusion Transformers

    Yonglak Son; Suhyeok Kim; Seungryong Kim; Young Geun Kim

    Paper ยท submitted 2025-12-30 ยท revised same as submission ยท cs.CV

    Method: CorGi Scope: image Unit: block/layer Reuse: direct reuse Training: training-free

  • [89] ProCache: Constraint-Aware Feature Caching with Selective Computation for Diffusion Transformer Acceleration

    Fanpu Cao; Yaofo Chen; Zeng You; Wei Luo

    Paper ยท submitted 2025-12-19 ยท revised 2026-02-13 ยท cs.CV

    Method: ProCache Scope: image Unit: feature Reuse: direct reuse

  • [88] No Cache Left Idle: Accelerating diffusion model via Extreme-slimming Caching

    Tingyan Wen; Haoyu Li; Yihuang Chen; Xing Zhou; Lifei Zhu; Xueqian Wang

    Paper ยท submitted 2025-12-14 ยท revised same as submission ยท cs.CV

    Method: ESC Scope: image Reuse: direct reuse

  • [87] REST: Diffusion-based Real-time End-to-end Streaming Talking Head Generation via ID-Context Caching and Asynchronous Streaming Distillation

    Haotian Wang; Yuzhe Weng; Jun Du; Haoran Xu; Xiaoyan Wu; Shan He et al.

    Paper ยท submitted 2025-12-12 ยท revised 2026-01-29 ยท cs.CV

    Method: REST Scope: video Reuse: learned approximation Reuse: direct reuse Training: learned

  • [86] InvarDiff: Cross-Scale Invariance Caching for Accelerated Diffusion Models

    Zihao Wu

    Paper ยท submitted 2025-11-29 ยท revised same as submission ยท cs.CV

    Method: InvarDiff Scope: image Reuse: direct reuse

  • [85] Fast3Dcache: Training-free 3D Geometry Synthesis Acceleration

    Mengyu Yang; Yanming Yang; Chenyi Xu; Chenxi Song; Yufan Zuo; Tong Zhao et al.

    Paper ยท Code ยท submitted 2025-11-27 ยท revised 2026-03-25 ยท cs.CV

    Method: Fast3Dcache Scope: 3D Unit: feature Reuse: forecasting Reuse: direct reuse Policy: similarity/geometry-aware Policy: adaptive Training: training-free

  • [84] H2-Cache: A Novel Hierarchical Dual-Stage Cache for High-Performance Acceleration of Generative Diffusion Models

    Mingyu Sung; Il-Min Kim; Sangseok Yun; Jae-Mo Kang

    Paper ยท Code ยท submitted 2025-10-31 ยท revised same as submission ยท cs.CV

    Method: H2-Cache Scope: image Reuse: direct reuse

  • [83] LeMiCa: Lexicographic Minimax Path Caching for Efficient Diffusion-Based Video Generation

    Huanlin Gao; Ping Chen; Fuyuan Shi; Chao Tan; Zhaoxiang Liu; Fang Zhao et al.

    Paper ยท Code ยท submitted 2025-10-30 ยท revised 2025-12-11 ยท cs.CV

    Method: LeMiCa Scope: video Reuse: direct reuse

  • [82] 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 et al.

    Paper ยท submitted 2025-10-22 ยท revised 2025-11-01 ยท cs.LG

    Method: โ€” Scope: image Reuse: direct reuse

  • [81] Attention Is All You Need for KV Cache in Diffusion LLMs

    Quan Nguyen-Tri; Mukul Ranjan; Zhiqiang Shen

    Paper ยท submitted 2025-10-16 ยท revised 2025-12-28 ยท cs.CL

    Method: โ€” Scope: diffusion LLM Unit: attention/KV Reuse: direct reuse

  • [80] Mask Tokens as Prophet: Fine-Grained Cache Eviction for Efficient dLLM Inference

    Jianuo Huang; Yaojie Zhang; Yicun Yang; Benhao Huang; Biqing Qi; Dongrui Liu et al.

    Paper ยท Code ยท submitted 2025-10-10 ยท revised same as submission ยท cs.CL

    Method: Mask Tokens as Prophet Scope: diffusion LLM Unit: token Reuse: direct reuse

  • [79] FreqCa: Accelerating Diffusion Models via Frequency-Aware Caching

    Jiacheng Liu; Peiliang Cai; Qinming Zhou; Yuqi Lin; Deyang Kong; Benhao Huang et al.

    Paper ยท submitted 2025-10-09 ยท revised same as submission ยท cs.LG

    Method: FreqCa Scope: image Reuse: direct reuse Policy: frequency-aware

  • [78] Predictive Feature Caching for Training-free Acceleration of Molecular Geometry Generation

    Johanna Sommer; John Rachwan; Nils Fleischmann; Stephan Gรผnnemann; Bertrand Charpentier

    Paper ยท submitted 2025-10-06 ยท revised 2026-06-22 ยท cs.LG

    Method: โ€” Scope: 3D Scope: scientific Unit: feature Reuse: forecasting Reuse: direct reuse Training: training-free

  • [77] Let Features Decide Their Own Solvers: Hybrid Feature Caching for Diffusion Transformers

    Shikang Zheng; Guantao Chen; Qinming Zhou; Yuqi Lin; Lixuan He; Chang Zou et al.

    Paper ยท submitted 2025-10-05 ยท revised same as submission ยท cs.CV

    Method: HyCa Scope: image Unit: feature Reuse: direct reuse

  • [76] d2^2Cache: Accelerating Diffusion-Based LLMs via Dual Adaptive Caching

    Yuchu Jiang; Yue Cai; Xiangzhong Luo; Jiale Fu; Jiarui Wang; Chonghan Liu et al.

    Paper ยท Code ยท submitted 2025-09-27 ยท revised 2026-02-16 ยท cs.CL

    Method: d2Cache Scope: diffusion LLM Reuse: direct reuse Policy: adaptive

  • [75] SpecDiff: Accelerating Diffusion Model Inference with Self-Speculation

    Jiayi Pan; Jiaming Xu; Yongkang Zhou; Guohao Dai

    Paper ยท submitted 2025-09-17 ยท revised 2025-11-22 ยท cs.CV

    Method: SpecDiff Scope: image

  • [74] BWCache: Accelerating Video Diffusion Transformers through Block-Wise Caching

    Hanshuai Cui; Zhiqing Tang; Zhifei Xu; Zhi Yao; Wenyi Zeng; Weijia Jia

    Paper ยท submitted 2025-09-17 ยท revised 2026-02-28 ยท cs.CV

    Method: BWCache Scope: video Unit: block/layer Reuse: direct reuse

  • [73] SpeCa: Accelerating Diffusion Transformers with Speculative Feature Caching

    Jiacheng Liu; Chang Zou; Yuanhuiyi Lyu; Fei Ren; Shaobo Wang; Kaixin Li et al.

    Paper ยท Code ยท DOI ยท submitted 2025-09-15 ยท revised same as submission ยท cs.LG

    Method: SpeCa Scope: image Unit: feature Reuse: direct reuse

  • [72] Compute Only 16 Tokens in One Timestep: Accelerating Diffusion Transformers with Cluster-Driven Feature Caching

    Zhixin Zheng; Xinyu Wang; Chang Zou; Shaobo Wang; Linfeng Zhang

    Paper ยท Code ยท submitted 2025-09-12 ยท revised same as submission ยท cs.CV

    Method: โ€” Scope: image Unit: token Unit: feature Unit: step Reuse: direct reuse Policy: similarity/geometry-aware

  • [71] DiTReducio: A Training-Free Acceleration for DiT-Based TTS via Progressive Calibration

    Yanru Huo; Ziyue Jiang; Zuoli Tang; Qingyang Hong; Zhou Zhao

    Paper ยท submitted 2025-09-11 ยท revised same as submission ยท cs.SD

    Method: DiTReducio Scope: audio Reuse: calibration/correction Training: training-free

  • [70] Accelerating Diffusion Transformer-Based Text-to-Speech with Transformer Layer Caching

    Siratish Sakpiboonchit

    Paper ยท submitted 2025-09-10 ยท revised same as submission ยท eess.AS

    Method: TTS Layer Cache Scope: audio Unit: block/layer Reuse: direct reuse

  • [69] Reusing Computation in Text-to-Image Diffusion for Efficient Generation of Image Sets

    Dale Decatur; Thibault Groueix; Wang Yifan; Rana Hanocka; Vladimir Kim; Matheus Gadelha

    Paper ยท submitted 2025-08-28 ยท revised same as submission ยท cs.CV

    Method: โ€” Scope: image

  • [68] Attacks on Approximate Caches in Text-to-Image Diffusion Models

    Desen Sun; Shuncheng Jie; Sihang Liu

    Paper ยท submitted 2025-08-28 ยท revised 2026-05-25 ยท cs.CR

    Method: โ€” Scope: image Reuse: direct reuse

  • [67] ERTACache: Error Rectification and Timesteps Adjustment for Efficient Diffusion

    Xurui Peng; Chenqian Yan; Hong Liu; Rui Ma; Fangmin Chen; Xing Wang et al.

    Paper ยท Code ยท submitted 2025-08-27 ยท revised 2026-02-10 ยท cs.CV

    Method: ERTACache Scope: image Unit: step Reuse: calibration/correction Reuse: direct reuse Policy: error/sensitivity-aware

  • [66] Lightning Fast Caching-based Parallel Denoising Prediction for Accelerating Talking Head Generation

    Jianzhi Long; Wenhao Sun; Rongcheng Tu; Dacheng Tao

    Paper ยท submitted 2025-08-25 ยท revised 2026-01-19 ยท cs.GR

    Method: LightningCP Scope: video Unit: model output Reuse: direct reuse Policy: system-aware

  • [65] HERO: Hierarchical Extrapolation and Refresh for Efficient World Models

    Quanjian Song; Xinyu Wang; Donghao Zhou; Jingyu Lin; Cunjian Chen; Yue Ma

    Paper ยท submitted 2025-08-25 ยท revised 2026-05-14 ยท cs.CV

    Method: HERO Scope: world model Reuse: forecasting

  • [64] DiCache: Let Diffusion Model Determine Its Own Cache

    Jiazi Bu; Pengyang Ling; Yujie Zhou; Yibin Wang; Yuhang Zang; Dahua Lin et al.

    Paper ยท Code ยท submitted 2025-08-24 ยท revised 2025-10-02 ยท cs.CV

    Method: DiCache Scope: image Reuse: direct reuse

  • [63] HiCache: A Plug-in Scaled-Hermite Upgrade for Taylor-Style Cache-then-Forecast Diffusion Acceleration

    Liang Feng; Shikang Zheng; Jiacheng Liu; Yuqi Lin; Qinming Zhou; Peiliang Cai et al.

    Paper ยท Code ยท submitted 2025-08-23 ยท revised 2026-01-26 ยท cs.CV

    Method: HiCache Scope: image Reuse: forecasting Reuse: direct reuse

    • HiCache++ (software; not a paper) โ€” Dynamic Mode Decomposition / Prony exponential feature forecasting with PyPI packaging and flow-matching 3D integrations for Hunyuan3D, SAM 3D, and TRELLIS. Its current results show a domain split: DMD performs best on the tested flow-matching 3D generators, while polynomial Taylor/Hermite bases remain preferable on DiT-XL/2 ImageNet-256.

      GitHub ยท PyPI ยท Zenodo DOI ยท Preprint

      Version: 1.2.1 Basis: DMD / Prony Scope: 3D generation License: MIT

      Maintainer: Krishi Attri ยท Release: 2026-07-02 ยท Software DOI: 10.5281/zenodo.21122835

  • [62] OmniCache: A Trajectory-Oriented Global Perspective on Training-Free Cache Reuse for Diffusion Transformer Models

    Huanpeng Chu; Wei Wu; Guanyu Fen; Yutao Zhang

    Paper ยท submitted 2025-08-22 ยท revised 2025-08-25 ยท cs.CV

    Method: OmniCache Scope: image Reuse: direct reuse Training: training-free

  • [61] Forecast then Calibrate: Feature Caching as ODE for Efficient Diffusion Transformers

    Shikang Zheng; Liang Feng; Xinyu Wang; Qinming Zhou; Peiliang Cai; Chang Zou et al.

    Paper ยท Code ยท submitted 2025-08-22 ยท revised same as submission ยท cs.CV

    Method: Forecast then Calibrate Scope: image Unit: feature Reuse: forecasting Reuse: calibration/correction Reuse: direct reuse

  • [60] Adaptive Hybrid Caching for Efficient Text-to-Video Diffusion Model Acceleration

    Yuanxin Wei; Lansong Diao; Bujiao Chen; Shenggan Cheng; Zhengping Qian; Wenyuan Yu et al.

    Paper ยท submitted 2025-08-18 ยท revised 2026-02-26 ยท cs.GR

    Method: MixCache Scope: video Reuse: direct reuse Policy: adaptive

  • [59] TaoCache: Structure-Maintained Video Generation Acceleration

    Zhentao Fan; Zongzuo Wang; Weiwei Zhang

    Paper ยท submitted 2025-08-12 ยท revised same as submission ยท cs.CV

    Method: TaoCache Scope: video Reuse: direct reuse

  • [58] Sparse-dLLM: Accelerating Diffusion LLMs with Dynamic Cache Eviction

    Yuerong Song; Xiaoran Liu; Ruixiao Li; Zhigeng Liu; Zengfeng Huang; Qipeng Guo et al.

    Paper ยท Code ยท submitted 2025-08-04 ยท revised 2025-11-05 ยท cs.CL

    Method: Sparse-dLLM Scope: diffusion LLM Reuse: direct reuse Policy: adaptive

  • [57] Sortblock: Similarity-Aware Feature Reuse for Diffusion Model

    Hanqi Chen; Xu Zhang; Xiaoliu Guan; Lielin Jiang; Guanzhong Wang; Zeyu Chen et al.

    Paper ยท submitted 2025-08-01 ยท revised 2026-01-07 ยท cs.CV

    Method: Sortblock Scope: image Unit: feature Reuse: direct reuse Policy: similarity/geometry-aware

  • [56] Less is Enough: Training-Free Video Diffusion Acceleration via Runtime-Adaptive Caching

    Xin Zhou; Dingkang Liang; Kaijin Chen; Tianrui Feng; Xiwu Chen; Hongkai Lin et al.

    Paper ยท Code ยท submitted 2025-07-03 ยท revised same as submission ยท cs.CV

    Method: Less is Enough Scope: video Reuse: direct reuse Policy: adaptive Training: training-free

  • [55] Evolutionary Caching to Accelerate Your Off-the-Shelf Diffusion Model

    Anirud Aggarwal; Abhinav Shrivastava; Matthew Gwilliam

    Paper ยท Code ยท submitted 2025-06-18 ยท revised 2026-03-02 ยท cs.CV

    Method: ECAD Scope: image Reuse: direct reuse Policy: budget/search

  • [54] Block-wise Adaptive Caching for Accelerating Diffusion Policy

    Kangye Ji; Yuan Meng; Hanyun Cui; Ye Li; Jianbo Zhou; Shengjia Hua et al.

    Paper ยท submitted 2025-06-16 ยท revised 2026-05-13 ยท cs.AI

    Method: BACache Scope: policy Unit: block/layer Reuse: direct reuse Policy: adaptive

  • [53] ReFrame: Layer Caching for Accelerated Inference in Real-Time Rendering

    Lufei Liu; Tor M. Aamodt

    Paper ยท submitted 2025-06-14 ยท revised same as submission ยท cs.GR

    Method: ReFrame Scope: image Unit: block/layer Reuse: direct reuse

  • [52] MagCache: Fast Video Generation with Magnitude-Aware Cache

    Zehong Ma; Longhui Wei; Feng Wang; Shiliang Zhang; Qi Tian

    Paper ยท Code ยท submitted 2025-06-10 ยท revised 2025-11-05 ยท cs.CV

    Method: MagCache Scope: video Reuse: direct reuse Policy: magnitude-aware

    Published reference: In Proceedings of NeurIPS 2025

  • [51] Chipmunk: Training-Free Acceleration of Diffusion Transformers with Dynamic Column-Sparse Deltas

    Austin Silveria; Soham V. Govande; Daniel Y. Fu

    Paper ยท Code ยท submitted 2025-06-03 ยท revised same as submission ยท cs.CV

    Method: Chipmunk Scope: image Reuse: delta reuse Policy: adaptive Training: training-free

  • [50] Foresight: Adaptive Layer Reuse for Accelerated and High-Quality Text-to-Video Generation

    Muhammad Adnan; Nithesh Kurella; Akhil Arunkumar; Prashant J. Nair

    Paper ยท Code ยท submitted 2025-05-31 ยท revised 2025-09-22 ยท cs.LG

    Method: Foresight Scope: video Unit: block/layer Reuse: direct reuse Policy: adaptive

  • [49] Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding

    Chengyue Wu; Hao Zhang; Shuchen Xue; Zhijian Liu; Shizhe Diao; Ligeng Zhu et al.

    Paper ยท submitted 2025-05-28 ยท revised 2025-07-03 ยท cs.CL

    Method: Fast-dLLM Scope: diffusion LLM Unit: attention/KV Reuse: direct reuse Policy: system-aware Training: training-free

  • [48] RainFusion: Adaptive Video Generation Acceleration via Multi-Dimensional Visual Redundancy

    Aiyue Chen; Bin Dong; Jingru Li; Jing Lin; Kun Tian; Yiwu Yao et al.

    Paper ยท submitted 2025-05-27 ยท revised 2025-06-09 ยท cs.CV

    Method: RainFusion Scope: video Policy: adaptive

  • [47] InstGenIE: Generative Image Editing Made Efficient with Mask-aware Caching and Scheduling

    Xiaoxiao Jiang; Suyi Li; Lingyun Yang; Tianyu Feng; Zhipeng Di; Weiyi Lu et al.

    Paper ยท submitted 2025-05-27 ยท revised same as submission ยท cs.DC

    Method: InstGenIE Scope: editing Reuse: direct reuse

  • [46] FlashDLM: Accelerating Diffusion Language Model Inference via Efficient KV Caching and Guided Diffusion

    Zhanqiu Hu; Jian Meng; Yash Akhauri; Mohamed S. Abdelfattah; Jae-sun Seo; Zhiru Zhang et al.

    Paper ยท submitted 2025-05-27 ยท revised 2025-10-09 ยท cs.CL

    Method: FlashDLM Scope: diffusion LLM Reuse: direct reuse

  • [45] FastCache: Fast Caching for Diffusion Transformer Through Learnable Linear Approximation

    Dong Liu; Yanxuan Yu; Jiayi Zhang; Yifan Li; Ben Lengerich; Ying Nian Wu

    Paper ยท Code ยท submitted 2025-05-26 ยท revised 2026-03-27 ยท cs.LG

    Method: FastCache Scope: image Reuse: learned approximation Reuse: direct reuse Training: learned

  • [44] dKV-Cache: The Cache for Diffusion Language Models

    Xinyin Ma; Runpeng Yu; Gongfan Fang; Xinchao Wang

    Paper ยท submitted 2025-05-21 ยท revised same as submission ยท cs.CL

    Method: dKV-Cache Scope: diffusion LLM Unit: attention/KV Reuse: direct reuse

  • [43] Communication-Efficient Diffusion Denoising Parallelization via Reuse-then-Predict Mechanism

    Kunyun Wang; Bohan Li; Kai Yu; Minyi Guo; Jieru Zhao

    Paper ยท submitted 2025-05-20 ยท revised 2025-10-13 ยท cs.LG

    Method: โ€” Scope: image Reuse: direct reuse Policy: system-aware Training: system

  • [42] CacheFlow: Fast Human Motion Prediction by Cached Normalizing Flow

    Takahiro Maeda; Jinkun Cao; Norimichi Ukita; Kris Kitani

    Paper ยท Code ยท submitted 2025-05-19 ยท revised 2026-01-30 ยท cs.CV

    Method: CacheFlow Scope: motion Reuse: direct reuse

    Published reference: Transactions on Machine Learning Research, 2026

  • [41] dLLM-Cache: Accelerating Diffusion Large Language Models with Adaptive Caching

    Zhiyuan Liu; Yicun Yang; Yaojie Zhang; Junjie Chen; Chang Zou; Qingyan Wei et al.

    Paper ยท Code ยท submitted 2025-05-17 ยท revised 2026-06-06 ยท cs.LG

    Method: dLLM-Cache Scope: diffusion LLM Reuse: direct reuse Policy: adaptive

  • [40] Accelerating Diffusion Transformer via Increment-Calibrated Caching with Channel-Aware Singular Value Decomposition

    Zhiyuan Chen; Keyi Li; Yifan Jia; Le Ye; Yufei Ma

    Paper ยท Code ยท submitted 2025-05-09 ยท revised same as submission ยท cs.CV

    Method: ICC Scope: image Reuse: calibration/correction Reuse: delta reuse Reuse: direct reuse

  • [39] AB-Cache: Training-Free Acceleration of Diffusion Models via Adams-Bashforth Cached Feature Reuse

    Zichao Yu; Zhen Zou; Guojiang Shao; Chengwei Zhang; Shengze Xu; Jie Huang et al.

    Paper ยท submitted 2025-04-13 ยท revised same as submission ยท stat.ML

    Method: AB-Cache Scope: image Unit: feature Reuse: direct reuse Training: training-free

  • [38] Model Reveals What to Cache: Profiling-Based Feature Reuse for Video Diffusion Models

    Xuran Ma; Yexin Liu; Yaofu Liu; Xianfeng Wu; Mingzhe Zheng; Zihao Wang et al.

    Paper ยท Code ยท submitted 2025-04-04 ยท revised 2025-07-26 ยท cs.CV

    Method: ProfilingDiT Scope: video Unit: feature Reuse: direct reuse

  • [37] BlockDance: Reuse Structurally Similar Spatio-Temporal Features to Accelerate Diffusion Transformers

    Hui Zhang; Tingwei Gao; Jie Shao; Zuxuan Wu

    Paper ยท submitted 2025-03-20 ยท revised same as submission ยท cs.CV

    Method: BlockDance Scope: image Unit: feature Reuse: direct reuse

  • [36] MoDM: Efficient Serving for Image Generation via Mixture-of-Diffusion Models

    Yuchen Xia; Divyam Sharma; Yichao Yuan; Souvik Kundu; Nishil Talati

    Paper ยท Code ยท submitted 2025-03-15 ยท revised 2025-08-02 ยท cs.DC

    Method: MoDM Scope: image Policy: system-aware Training: system

  • [35] From Reusing to Forecasting: Accelerating Diffusion Models with TaylorSeers

    Jiacheng Liu; Chang Zou; Yuanhuiyi Lyu; Junjie Chen; Linfeng Zhang

    Paper ยท Code ยท submitted 2025-03-10 ยท revised 2025-08-11 ยท cs.CV

    Method: From Reusing to Forecasting Scope: image Reuse: forecasting

  • [34] FEB-Cache: Frequency-Guided Exposure Bias Reduction for Enhancing Diffusion Transformer Caching

    Zhen Zou; Feng Zhao

    Paper ยท Code ยท submitted 2025-03-10 ยท revised 2025-10-06 ยท cs.CV

    Method: FEB-Cache Scope: image Reuse: direct reuse Policy: frequency-aware

  • [33] QuantCache: Adaptive Importance-Guided Quantization with Hierarchical Latent and Layer Caching for Video Generation

    Junyi Wu; Zhiteng Li; Zheng Hui; Yulun Zhang; Linghe Kong; Xiaokang Yang

    Paper ยท Code ยท submitted 2025-03-09 ยท revised same as submission ยท cs.CV

    Method: QuantCache Scope: video Unit: block/layer Unit: latent Reuse: compression Reuse: direct reuse Policy: adaptive

  • [32] Accelerating Diffusion Transformer via Gradient-Optimized Cache

    Junxiang Qiu; Lin Liu; Shuo Wang; Jinda Lu; Kezhou Chen; Yanbin Hao

    Paper ยท Code ยท submitted 2025-03-07 ยท revised 2025-07-18 ยท cs.CV

    Method: GOC Scope: image Reuse: direct reuse

    Published reference: ICCV 2025

  • [31] Q&C: When Quantization Meets Cache in Efficient Image Generation

    Xin Ding; Xin Li; Haotong Qin; Zhibo Chen

    Paper ยท Code ยท submitted 2025-03-04 ยท revised same as submission ยท cs.CV

    Method: Q&C Scope: image Reuse: compression Reuse: direct reuse

  • [30] CacheQuant: Comprehensively Accelerated Diffusion Models

    Xuewen Liu; Zhikai Li; Qingyi Gu

    Paper ยท Code ยท submitted 2025-03-03 ยท revised same as submission ยท cs.CV

    Method: CacheQuant Scope: image Reuse: direct reuse

  • [29] UniCP: A Unified Caching and Pruning Framework for Efficient Video Generation

    Wenzhang Sun; Qirui Hou; Donglin Di; Jiahui Yang; Yongjia Ma; Jianxun Cui

    Paper ยท submitted 2025-02-06 ยท revised same as submission ยท cs.CV

    Method: UniCP Scope: video Reuse: direct reuse

  • [28] Accelerating Diffusion Transformer via Error-Optimized Cache

    Junxiang Qiu; Shuo Wang; Jinda Lu; Lin Liu; Houcheng Jiang; Xingyu Zhu et al.

    Paper ยท Code ยท submitted 2025-01-31 ยท revised 2025-07-18 ยท cs.CV

    Method: EOC Scope: image Reuse: direct reuse Policy: error/sensitivity-aware

    Published reference: ACM MM 2025

  • [27] Cached Adaptive Token Merging: Dynamic Token Reduction and Redundant Computation Elimination in Diffusion Model

    Omid Saghatchian; Atiyeh Gh. Moghadam; Ahmad Nickabadi

    Paper ยท submitted 2025-01-01 ยท revised same as submission ยท cs.CV

    Method: โ€” Scope: image Unit: token Reuse: direct reuse Policy: adaptive

๐Ÿ—“ 2024

2024 ยท 20 papers

  • [26] Token Pruning for Caching Better: 9 Times Acceleration on Stable Diffusion for Free

    Evelyn Zhang; Bang Xiao; Jiayi Tang; Qianli Ma; Chang Zou; Xuefei Ning et al.

    Paper ยท submitted 2024-12-31 ยท revised same as submission ยท cs.CV

    Method: โ€” Scope: image Unit: token Reuse: direct reuse

  • [25] Rethinking Token-wise Feature Caching: Accelerating Diffusion Transformers with Dual Feature Caching

    Chang Zou; Evelyn Zhang; Runlin Guo; Haohang Xu; Conghui He; Xuming Hu et al.

    Paper ยท Code ยท submitted 2024-12-25 ยท revised 2025-11-18 ยท cs.LG

    Method: DuCa Scope: image Unit: token Unit: feature Reuse: direct reuse

  • [24] FlexCache: Flexible Approximate Cache System for Video Diffusion

    Desen Sun; Henry Tian; Tim Lu; Sihang Liu

    Paper ยท submitted 2024-12-18 ยท revised same as submission ยท cs.MM

    Method: FlexCache Scope: video Reuse: direct reuse Policy: system-aware Training: system

  • [23] LazyDiT: Lazy Learning for the Acceleration of Diffusion Transformers

    Xuan Shen; Zhao Song; Yufa Zhou; Bo Chen; Yanyu Li; Yifan Gong et al.

    Paper ยท Code ยท submitted 2024-12-17 ยท revised 2025-03-21 ยท cs.LG

    Method: LazyDiT Scope: image

  • [22] Timestep Embedding Tells: It's Time to Cache for Video Diffusion Model

    Feng Liu; Shiwei Zhang; Xiaofeng Wang; Yujie Wei; Haonan Qiu; Yuzhong Zhao et al.

    Paper ยท Code ยท submitted 2024-11-28 ยท revised 2025-03-18 ยท cs.CV

    Method: Timestep Embedding Tells Scope: video Unit: step Reuse: direct reuse

  • [21] Towards Stabilized and Efficient Diffusion Transformers through Long-Skip-Connections with Spectral Constraints

    Guanjie Chen; Xinyu Zhao; Yucheng Zhou; Xiaoye Qu; Tianlong Chen; Yu Cheng

    Paper ยท Code ยท submitted 2024-11-26 ยท revised 2026-02-07 ยท cs.CV

    Method: SkipCache Scope: image Reuse: direct reuse Policy: frequency-aware

  • [20] Ca2-VDM: Efficient Autoregressive Video Diffusion Model with Causal Generation and Cache Sharing

    Kaifeng Gao; Jiaxin Shi; Hanwang Zhang; Chunping Wang; Jun Xiao; Long Chen

    Paper ยท Code ยท submitted 2024-11-25 ยท revised 2025-05-21 ยท cs.CV

    Method: Ca2-VDM Scope: video Reuse: direct reuse

  • [19] SmoothCache: A Universal Inference Acceleration Technique for Diffusion Transformers

    Joseph Liu; Joshua Geddes; Ziyu Guo; Haomiao Jiang; Mahesh Kumar Nandwana

    Paper ยท Code ยท submitted 2024-11-15 ยท revised 2025-05-21 ยท cs.LG

    Method: SmoothCache Scope: image Reuse: direct reuse

  • [18] FasterCache: Training-Free Video Diffusion Model Acceleration with High Quality

    Zhengyao Lv; Chenyang Si; Junhao Song; Zhenyu Yang; Yu Qiao; Ziwei Liu et al.

    Paper ยท Code ยท submitted 2024-10-25 ยท revised 2025-03-12 ยท cs.CV

    Method: FasterCache Scope: video Reuse: direct reuse Training: training-free

  • [17] Accelerating Diffusion Transformers with Token-wise Feature Caching

    Chang Zou; Xuyang Liu; Ting Liu; Siteng Huang; Linfeng Zhang

    Paper ยท Code ยท submitted 2024-10-05 ยท revised 2025-02-19 ยท cs.LG

    Method: ToCa Scope: image Unit: token Unit: feature Reuse: direct reuse

  • [16] HarmoniCa: Harmonizing Training and Inference for Better Feature Caching in Diffusion Transformer Acceleration

    Yushi Huang; Zining Wang; Ruihao Gong; Jing Liu; Xinjie Zhang; Jinyang Guo et al.

    Paper ยท Code ยท submitted 2024-10-02 ยท revised 2025-05-31 ยท cs.CV

    Method: HarmoniCa Scope: image Unit: feature Reuse: direct reuse

  • [15] Token Caching for Diffusion Transformer Acceleration

    Jinming Lou; Wenyang Luo; Yufan Liu; Bing Li; Xinmiao Ding; Weiming Hu et al.

    Paper ยท submitted 2024-09-27 ยท revised 2026-01-27 ยท cs.LG

    Method: TokenCache Scope: image Unit: token Reuse: direct reuse

  • [14] Real-Time Video Generation with Pyramid Attention Broadcast

    Xuanlei Zhao; Xiaolong Jin; Kai Wang; Yang You

    Paper ยท Code ยท submitted 2024-08-22 ยท revised 2025-02-27 ยท cs.CV

    Method: PAB Scope: video Unit: attention/KV Reuse: direct reuse

  • [13] Faster Image2Video Generation: A Closer Look at CLIP Image Embedding's Impact on Spatio-Temporal Cross-Attentions

    Ashkan Taghipour; Morteza Ghahremani; Mohammed Bennamoun; Aref Miri Rekavandi; Zinuo Li; Hamid Laga et al.

    Paper ยท submitted 2024-07-27 ยท revised same as submission ยท cs.CV

    Method: VCUT Scope: video Unit: attention/KV

  • [12] FORA: Fast-Forward Caching in Diffusion Transformer Acceleration

    Pratheba Selvaraju; Tianyu Ding; Tianyi Chen; Ilya Zharkov; Luming Liang

    Paper ยท Code ยท submitted 2024-07-01 ยท revised same as submission ยท cs.CV

    Method: FORA Scope: image Reuse: direct reuse

  • [11] DiTFastAttn: Attention Compression for Diffusion Transformer Models

    Zhihang Yuan; Hanling Zhang; Pu Lu; Xuefei Ning; Linfeng Zhang; Tianchen Zhao et al.

    Paper ยท Code ยท submitted 2024-06-12 ยท revised 2024-10-18 ยท cs.CV

    Method: DiTFastAttn Scope: image Reuse: compression

  • [10] Learning-to-Cache: Accelerating Diffusion Transformer via Layer Caching

    Xinyin Ma; Gongfan Fang; Michael Bi Mi; Xinchao Wang

    Paper ยท Code ยท submitted 2024-06-03 ยท revised 2024-11-16 ยท cs.LG

    Method: Learning-to-Cache Scope: image Unit: block/layer Reuse: learned approximation Reuse: direct reuse Training: learned

  • [9] ฮ”ฮ”-DiT: A Training-Free Acceleration Method Tailored for Diffusion Transformers

    Pengtao Chen; Mingzhu Shen; Peng Ye; Jianjian Cao; Chongjun Tu; Christos-Savvas Bouganis et al.

    Paper ยท submitted 2024-06-03 ยท revised same as submission ยท cs.CV

    Method: ฮ”-DiT Scope: image Training: training-free

  • [8] Hash3D: Training-free Acceleration for 3D Generation

    Xingyi Yang; Xinchao Wang

    Paper ยท Code ยท submitted 2024-04-09 ยท revised same as submission ยท cs.CV

    Method: Hash3D Scope: 3D Unit: feature Reuse: direct reuse Policy: similarity/geometry-aware Training: training-free

  • [7] Faster Diffusion via Temporal Attention Decomposition

    Haozhe Liu; Wentian Zhang; Jinheng Xie; Francesco Faccio; Mengmeng Xu; Tao Xiang et al.

    Paper ยท Code ยท submitted 2024-04-03 ยท revised 2025-02-26 ยท cs.CV

    Method: T-GATE Scope: image

๐Ÿ—“ 2023

2023 ยท 6 papers

  • [6] Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference

    Senmao Li; Taihang Hu; Joost van de Weijer; Fahad Shahbaz Khan; Tao Liu; Linxuan Li et al.

    Paper ยท Code ยท submitted 2023-12-15 ยท revised 2024-10-15 ยท cs.CV

    Method: FasterDiffusion Scope: image

  • [5] Approximate Caching for Efficiently Serving Diffusion Models

    Shubham Agarwal; Subrata Mitra; Sarthak Chakraborty; Srikrishna Karanam; Koyel Mukherjee; Shiv Saini

    Paper ยท submitted 2023-12-07 ยท revised same as submission ยท cs.CV

    Method: Approximate Caching Scope: image Reuse: direct reuse Policy: system-aware Training: system

  • [4] FRDiff : Feature Reuse for Universal Training-free Acceleration of Diffusion Models

    Junhyuk So; Jungwon Lee; Eunhyeok Park

    Paper ยท Code ยท submitted 2023-12-06 ยท revised 2024-09-02 ยท cs.CV

    Method: FRDiff Scope: image Unit: feature Reuse: direct reuse Training: training-free

  • [3] Cache Me if You Can: Accelerating Diffusion Models through Block Caching

    Felix Wimbauer; Bichen Wu; Edgar Schoenfeld; Xiaoliang Dai; Ji Hou; Zijian He et al.

    Paper ยท submitted 2023-12-06 ยท revised 2024-01-12 ยท cs.CV

    Method: Block Caching Scope: image Unit: block/layer Reuse: direct reuse

  • [2] DeepCache: Accelerating Diffusion Models for Free

    Xinyin Ma; Gongfan Fang; Xinchao Wang

    Paper ยท Code ยท submitted 2023-12-01 ยท revised 2023-12-07 ยท cs.CV

    Method: DeepCache Scope: image Reuse: direct reuse

  • [1] Accelerating Text-to-Image Editing via Cache-Enabled Sparse Diffusion Inference

    Zihao Yu; Haoyang Li; Fangcheng Fu; Xupeng Miao; Bin Cui

    Paper ยท Code ยท submitted 2023-05-27 ยท revised 2024-01-04 ยท cs.CV

    Method: FISEdit Scope: editing Reuse: direct reuse

๐Ÿค Contributing

Please read CONTRIBUTING.md. New entries need a public primary source and complete taxonomy fields.

๐Ÿ™ Acknowledgments

Maintained by EPIC Lab at Shanghai Jiao Tong University. Corrections and contributions are welcome through issues and pull requests.