cs.CVFeb 26, 2026

Denoising as Path Planning: Training-Free Acceleration of Diffusion Models with DPCache

Authors: Bowen Cui, Yuanbin Wang, Huajiang Xu, Biaolong Chen, Aixi Zhang, Hao Jiang, Zhengzheng Jin, Xu Liu, 

Organizations: Alibaba Group

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Diffusion models have demonstrated remarkable success in image and video generation, yet their practical deployment remains hindered by the substantial computational overhead of multi-step iterative sampling. Among acceleration strategies, caching-based methods offer a training-free and effective solution by reusing or predicting features across timesteps. However, existing approaches rely on fixed or locally adaptive schedules without considering the global structure of the denoising trajectory, often leading to error accumulation and visual artifacts. To overcome this limitation, we propose DPCache, a novel training-free acceleration framework that formulates diffusion sampling acceleration as a global path planning problem. DPCache constructs a Path-Aware Cost Tensor from a small calibration set to quantify the path-dependent error of skipping timesteps conditioned on the preceding key timestep. Leveraging this tensor, DPCache employs dynamic programming to select an optimal sequence of key timesteps that minimizes the total path cost while preserving trajectory fidelity. During inference, the model performs full computations only at these key timesteps, while intermediate outputs are efficiently predicted using cached features. Extensive experiments on DiT, FLUX, and HunyuanVideo demonstrate that DPCache achieves strong acceleration with minimal quality loss, outperforming prior acceleration methods by ++0.031 ImageReward at 4.87×\times speedup and even surpassing the full-step baseline by ++0.028 ImageReward at 3.54×\times speedup on FLUX, validating the effectiveness of our path-aware global scheduling framework. Code is available at https://github.com/argsss/DPCache.

Figures & tables

Figure 1: (a) Fixed schedule is inflexible and unable to identify critical timesteps, resulting in large deviation from the true trajectory. (b) Locally adaptive schedule makes greedy, short-sighted decisions that often skip essential timesteps, leading to irreversible deviation. (c) DPCache identifies a globally optimal sequence of key timesteps through calibration and achieves low cumulative trajectory deviation.
Figure 2: Overview of DPCache. (a) During the calibration stage, the full TT -step denoising process is executed to construct a 3D Path-Aware Cost Tensor (PACT), which quantifies the cumulative error of skipping intermediate timesteps conditioned on the preceding key step. (b) An optimal KK -step sampling schedule ( K<TK<T ) is selected via dynamic programming, leveraging the PACT to maintain a DP table to store minimal cumulative costs and a path table to enable backtracking. (c) During inference, the model computes and caches features only at the preselected key timesteps, while features at other timesteps are efficiently predicted.
Figure 3: Qualitative comparison of different acceleration methods applied to FLUX.1-dev on DrawBench dataset.
Figure 4: Qualitative comparison of different acceleration methods applied to HunyuanVideo on VBench dataset.
Appendix figures & tables9 assets

Supplementary material from the paper’s appendix.

Appendix

ModelMethodHyperparameters
FLUX.1-devTeaCache 1l=0.7l=0.7
TeaCache 2l=1.0l=1.0
TaylorSeer 1N=5,O=2\mathcal{N}=5,O=2
TaylorSeer 2N=8,O=2\mathcal{N}=8,O=2
SpeCa 1τ=1.0,β=0.2\tau=1.0,\beta=0.2
SpeCa 2τ=8.5,β=0.4\tau=8.5,\beta=0.4
Appendix
Table 6: Hyperparameters used for each method in our experiments.
Figure 5: Visualization of sampling trajectories when applying different acceleration methods to FLUX.1-dev.
Figure 6: VBench performance of HunyuanVideo across all dimensions with various acceleration methods under different acceleration settings: (left) low speedup, (right) high speedup. Scores are normalized per dimension for improved visual comparison.
Figure 7: Additional qualitative comparisons of different acceleration methods applied to FLUX.1-dev on DrawBench dataset.
Figure 8: Qualitative comparison between the original (unaccelerated) and DPCache-accelerated Wan-2.1-I2V-14B on the VBench-I2V benchmark.
Figure 9: Qualitative comparison of different acceleration methods applied to FLUX.1-dev on out-of-distribution prompts.
Figure 10: Visualization of ablation studies on the dimensionality and cost aggregation strategy of PACT.
Figure 11: Performance of DPCache under varying acceleration ratios and prediction orders.
Figure 12: Failure cases of DPCache.
More in this topic

Diffusion Model Inference Acceleration

Diffusion model inference acceleration reduces the time and compute needed to generate or reconstruct outputs while preserving quality.

See all 491 papers

Explore similar work

CardsList
Jul 31, 2026

OnlineCache: Learning Dynamic Caching Policies with Error Correction for Efficient Diffusion Inference

Diffusion models have revolutionized generative tasks but incur high latency due to iterative denoising. While cache-based strategies accelerate inference by reusing intermediate features, they largely rely on static, sample-agnostic schedules. We argue that this rigidity overlooks two facts empirically validated in this paper: (i) generation difficulty varies across prompts, requiring adaptive resource allocation--complex inputs demand more computation while simpler ones require less; (ii) error sensitivity fluctuates across timesteps, where static policies may cache high-error steps or waste computation on low-error ones. We therefore propose OnlineCache, a dynamic caching framework that jointly learns when to cache and how to correct approximation errors. We leverage policy gradient to train a lightweight network for adaptive speed-quality trade-offs, and incorporate a learnable corrector to mitigate caching-induced errors. Both modules are jointly optimized under a bilevel optimization framework, with the policy targeting global generation quality and the corrector minimizing local errors. Our method automatically allocates computational resources across both samples and timesteps, improving overall generation quality. Extensive experiments demonstrate clear superiority. On FLUX.1-dev model, OnlineCache achieves nearly 3 speedup while preserving generation fidelity. On DiT and CogVideoX, it similarly delivers competitive acceleration without compromising quality; across all scenarios, it consistently outperforms existing cache-based acceleration baselines.

Zhikang Xie, Xichen Ye, Yifan Wu +5

College of Computer Science and Artificial Intelligence, Fudan University · School of Data Science, Fudan University · 3Shanghai Key Laboratory of Intelligent Information Processing

Diffusion Model CachingDiffusion Model Inference AccelerationDiffusion Models
AskPDF
Mar 3, 2026

TC-Padé: Trajectory-Consistent Padé Approximation for Diffusion Acceleration

Despite achieving state-of-the-art generation quality, diffusion models are hindered by the substantial computational burden of their iterative sampling process. While feature caching techniques achieve effective acceleration at higher step counts (e.g., 50 steps), they exhibit critical limitations in the practical low-step regime of 20-30 steps. As the interval between steps increases, polynomial-based extrapolators like TaylorSeer suffer from error accumulation and trajectory drift. Meanwhile, conventional caching strategies often overlook the distinct dynamical properties of different denoising phases. To address these challenges, we propose Trajectory-Consistent Padé approximation, a feature prediction framework grounded in Padé approximation. By modeling feature evolution through rational functions, our approach captures asymptotic and transitional behaviors more accurately than Taylor-based methods. To enable stable and trajectory-consistent sampling under reduced step counts, TC-Padé incorporates (1) adaptive coefficient modulation that leverages historical cached residuals to detect subtle trajectory transitions, and (2) step-aware prediction strategies tailored to the distinct dynamics of early, mid, and late sampling stages. Extensive experiments on DiT-XL/2, FLUX.1-dev, and Wan2.1 across both image and video generation demonstrate the effectiveness of TC-Padé. For instance, TC-Padé achieves 2.88x acceleration on FLUX.1-dev and 1.72x on Wan2.1 while maintaining high quality across FID, CLIP, Aesthetic, and VBench-2.0 metrics, substantially outperforming existing feature caching methods.

Benlei Cui, Shaoxuan He, Bukun Huang +8

Yuvion Team, Alibaba Group · Zhejiang University · Fireflyfusion, QuVideo Inc +1

Diffusion Model AccelerationFew-Step Diffusion SamplingDiffusion Model CachingVideo Diffusion Models
AskPDF
Feb 20, 2026

Predict to Skip: Linear Multistep Feature Forecasting for Efficient Diffusion Transformers

Diffusion Transformers (DiT) have emerged as a widely adopted backbone for high-fidelity image and video generation, yet their iterative denoising process incurs high computational costs. Existing training-free acceleration methods rely on feature caching and reuse under the assumption of temporal stability. However, reusing features for multiple steps may lead to latent drift and visual degradation. We observe that model outputs evolve smoothly along much of the diffusion trajectory, enabling principled predictions rather than naive reuse. Based on this insight, we propose PrediT, a training-free acceleration framework that formulates feature prediction as a linear multistep problem. We employ classical linear multistep methods to forecast future model outputs from historical information, combined with a corrector that activates in high-dynamics regions to prevent error accumulation. A dynamic step modulation mechanism adaptively adjusts the prediction horizon by monitoring the feature change rate. Together, these components enable substantial acceleration while preserving generation fidelity. Extensive experiments validate that our method achieves up to 5.54×5.54\times latency reduction across various DiT-based image and video generation models, while incurring negligible quality degradation.

Hanshuai Cui, Zhiqing Tang, Qianli Ma +3

School of Artificial Intelligence, Beijing Normal University, Beijing, China · Institute of Artificial Intelligence and Future Networks, Beijing Normal University, Zhuhai, China · Shenzhen University of Advanced Technology, Shenzhen, China

Diffusion Model AccelerationDiffusion SamplingDiffusion-Based Image GenerationDiffusion TransformerVideo Diffusion Models
AskPDF