mlsanityData to 14 Oct 2026

Accelerating Diffusion Transformers with Token-wise Feature Caching

arXiv 2410.05317 ↗First posted 5 Oct 2024cs.LG cs.AI cs.CVcited by 156

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

Diffusion transformers have shown significant effectiveness in both image and video synthesis at the expense of huge computation costs. To address this problem, feature caching methods have been introduced to accelerate diffusion transformers by caching the features in previous timesteps and reusing them in the following timesteps. However, previous caching methods ignore that different tokens exhibit different sensitivities to feature caching, and feature caching on some tokens may lead to 10× more destruction to the overall generation quality compared with other tokens. In this paper, we introduce token-wise feature caching, allowing us to adaptively select the most suitable tokens for caching, and further enable us to apply different caching ratios to neural layers in different types and depths. Extensive experiments on PixArt-α, OpenSora, and DiT demonstrate our effectiveness in both image and video generation with no requirements for training. For instance, 2.36× and 1.93× acceleration are achieved on OpenSora and PixArt-α with almost no drop in generation quality.

Key facts. Cited by 156 papers in the map (data through 14 Oct 2026). Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models. Builds on 29 of its 39 references that are in the map. First posted on arXiv 5 Oct 2024.

Builds on

29 of 39 references.

Closest ancestors · 18

Same direction, newest first

  1. 2024FORA: Fast-Forward Caching in Diffusion Transformer Acceleration1 Jul
  2. 2024Learning-to-Cache: Accelerating Diffusion Transformer via Layer Caching3 Jun
  3. 2024Delta-DiT: A Training-Free Acceleration Method Tailored for Diffusion Transformers3 Jun
  4. 2023Cache Me if You Can: Accelerating Diffusion Models through Block Caching6 Dec
  5. 2023DeepCache: Accelerating Diffusion Models for Free1 Dec
  6. 2023PixArt-α: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis30 Sep
  7. 2023Structural Pruning for Diffusion Models18 May
  8. 2023Token Merging for Fast Stable Diffusion30 Mar
  9. 2023Consistency Models2 Mar
  10. 2023Q-Diffusion: Quantizing Diffusion Models8 Feb
  11. 2022Scalable Diffusion Models with Transformers19 Dec
  12. 2022Post-training Quantization on Diffusion Models28 Nov
6 more
  1. 2022DPM-Solver++: Fast Solver for Guided Sampling of Diffusion Probabilistic Models2 Nov
  2. 2022On Distillation of Guided Diffusion Models6 Oct
  3. 2022DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps2 Jun
  4. 2021High-Resolution Image Synthesis with Latent Diffusion Models20 Dec
  5. 2020Denoising Diffusion Implicit Models6 Oct
  6. 2020Denoising Diffusion Probabilistic Models19 Jun

Other references · 11

  1. 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,713
  2. 2014Microsoft COCO: Common Objects in Contextcited by 17,608
  3. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,058
  4. 2022Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flowcited by 3,700
  5. 2023Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasetscited by 2,581
  6. 2021CLIPScore: A Reference-free Evaluation Metric for Image Captioningcited by 2,424
  7. 2023VBench: Comprehensive Benchmark Suite for Video Generative Modelscited by 1,822
  8. 2022Scaling Autoregressive Models for Content-Rich Text-to-Image Generationcited by 1,295
  9. 2022eDiff-I: Text-to-Image Diffusion Models with an Ensemble of Expert Denoiserscited by 931
  10. 2023SnapFusion: Text-to-Image Diffusion Model on Mobile Devices within Two Secondscited by 227
  11. 2024Real-Time Video Generation with Pyramid Attention Broadcastcited by 149

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 2; 2025: 61; 2026: 93.

156 papers.

Sort:

By area:

  1. 2024From Slow Bidirectional to Fast Autoregressive Video Diffusion Modelscited by 433
  2. 2025dLLM-Cache: Accelerating Diffusion Large Language Models with Adaptive Cachingcited by 169
  3. 2025From Reusing to Forecasting: Accelerating Diffusion Models with TaylorSeerscited by 136
  4. 2025EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Modelscited by 93
  5. 2025Autoregressive Adversarial Post-Training for Real-Time Interactive Video Generationcited by 90
  6. 2025Yume: An Interactive World Generation Modelcited by 81
  7. 2025OminiControl2: Efficient Conditioning for Diffusion Transformerscited by 57
  8. 2025Token Reduction Should Go Beyond Efficiency in Generative Models -- From Vision, Language to Multimodalitycited by 50
  9. 2025EEdit: Rethinking the Spatial and Temporal Redundancy for Efficient Image Editingcited by 45
  10. 2025Less is Enough: Training-Free Video Diffusion Acceleration via Runtime-Adaptive Cachingcited by 45
  11. 2025Follow-Your-Emoji-Faster: Towards Efficient, Fine-Controllable, and Expressive Freestyle Portrait Animationcited by 43
  12. 2025DiCache: Let Diffusion Model Determine Its Own Cachecited by 42
38 more
  1. 2025FastVAR: Linear Visual Autoregressive Modeling via Cached Token Pruningcited by 34
  2. 2025Shifting AI Efficiency From Model-Centric to Data-Centric Compressioncited by 29
  3. 2025SpeCa: Accelerating Diffusion Transformers with Speculative Feature Cachingcited by 29
  4. 2024Rethinking Token-wise Feature Caching: Accelerating Diffusion Transformers with Dual Feature Cachingcited by 28
  5. 2025Training-free Diffusion Acceleration with Bottleneck Samplingcited by 25
  6. 2025QVGen: Pushing the Limit of Quantized Video Generative Modelscited by 22
  7. 2025FreqCa: Accelerating Diffusion Models via Frequency-Aware Cachingcited by 20
  8. 2025Compute Only 16 Tokens in One Timestep: Accelerating Diffusion Transformers with Cluster-Driven Feature Cachingcited by 19
  9. 2025OmniCache: A Trajectory-Oriented Global Perspective on Training-Free Cache Reuse for Diffusion Transformer Modelscited by 18
  10. 2025A Survey on Cache Methods in Diffusion Models: Toward Efficient Multi-Modal Generationcited by 16
  11. 2025Model Reveals What to Cache: Profiling-Based Feature Reuse for Video Diffusion Modelscited by 15
  12. 2025Accelerating Diffusion Transformer via Increment-Calibrated Caching with Channel-Aware Singular Value Decompositioncited by 12
  13. 2025HiCache: A Plug-in Scaled-Hermite Upgrade for Taylor-Style Cache-then-Forecast Diffusion Accelerationcited by 12
  14. 2025LazyMAR: Accelerating Masked Autoregressive Models via Feature Cachingcited by 11
  15. 2025Astraea: A Token-wise Acceleration Framework for Video Diffusion Transformerscited by 11
  16. 2025SkipVAR: Accelerating Visual Autoregressive Modeling via Adaptive Frequency-Aware Skippingcited by 11
  17. 2025SADA: Stability-guided Adaptive Diffusion Accelerationcited by 11
  18. 2026SeaCache: Spectral-Evolution-Aware Cache for Accelerating Diffusion Modelscited by 11
  19. 2025Training-free Mixed-Resolution Latent Upsampling for Spatially Accelerated Diffusion Transformerscited by 10
  20. 2025Let Features Decide Their Own Solvers: Hybrid Feature Caching for Diffusion Transformerscited by 10
  21. 2025LeMiCa: Lexicographic Minimax Path Caching for Efficient Diffusion-Based Video Generationcited by 10
  22. 2026MeanCache: From Instantaneous to Average Velocity for Accelerating Flow Matching Inferencecited by 10
  23. 2026TC-Pad'e: Trajectory-Consistent Pad'e Approximation for Diffusion Accelerationcited by 10
  24. 2025HERO: Hierarchical Extrapolation and Refresh for Efficient World Modelscited by 9
  25. 2025Chipmunk: Training-Free Acceleration of Diffusion Transformers with Dynamic Column-Sparse Deltascited by 8
  26. 2026Denoising as Path Planning: Training-Free Acceleration of Diffusion Models with DPCachecited by 8
  27. 2026DisCa: Accelerating Video Diffusion Transformers with Distillation-Compatible Learnable Feature Cachingcited by 7
  28. 2026Adaptive Spectral Feature Forecasting for Diffusion Sampling Accelerationcited by 7
  29. 2026Efficient Video Diffusion Models: Advancements and Challengescited by 7
  30. 2025Foresight: Adaptive Layer Reuse for Accelerated and High-Quality Text-to-Video Generationcited by 6
  31. 2025Revisiting Diffusion Models: From Generative Pre-training to One-Step Generationcited by 6
  32. 2025Forecasting When to Forecast: Accelerating Diffusion Models with Confidence-Gated Taylorcited by 6
  33. 2025RAPID^3: Tri-Level Reinforced Acceleration Policies for Diffusion Transformercited by 6
  34. 2025RegionE: Adaptive Region-Aware Generation for Efficient Image Editingcited by 6
  35. 2025FreeSwim: Revisiting Sliding-Window Attention Mechanisms for Training-Free Ultra-High-Resolution Video Generationcited by 6
  36. 2025TS-DP: Reinforcement Speculative Decoding For Temporal Adaptive Diffusion Policy Accelerationcited by 6
  37. 2026WorldCache: Accelerating World Models for Free via Heterogeneous Token Cachingcited by 6
  38. 2026ZEUS: Accelerating Diffusion Models with Only Second-Order Predictorcited by 6

50 most cited of 156 All 156 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024Token Caching for Diffusion Transformer Acceleration
  2. 2025Forecast then Calibrate: Feature Caching as ODE for Efficient Diffusion Transformers
  3. 2024HarmoniCa: Harmonizing Training and Inference for Better Feature Caching in Diffusion Transformer Acceleration
  4. 2025Accelerating Diffusion Transformer via Gradient-Optimized Cache
  5. 2025BWCache: Accelerating Video Diffusion Transformers through Block-Wise Caching
  6. 2025FastCache: Fast Caching for Diffusion Transformer Through Learnable Linear Approximation
  7. 2025Accelerating Diffusion Transformer via Error-Optimized Cache
  8. 2024Towards Stabilized and Efficient Diffusion Transformers through Long-Skip-Connections with Spectral Constraints