NaviCache: Test-Time Self-Calibration Caching for Video Generation

Zheqi Lv, Zhibo Zhu, Jinke Wang, Qi Tian, Shengyu Zhang, Zhengyu Chen, Chengxi Zang, Zhou Zhao, Fei Wu
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:83724-83737, 2026.

Abstract

Video Diffusion Models (VDMs) is constrained by immense computational costs. While offline calibration-based acceleration suffers from calibration data dependency, prohibitive calibration duration, and susceptibility to distribution shifts, offline calibration-free methods eliminate these hurdles. However, since they rely on instantaneous zero-order approximations where the mapping between input and output differences varies in real-time, they are susceptible to observational noise and ignore the intrinsic momentum within the diffusion trajectory. In this paper, we propose NaviCache, a plug-and-play test-time self-calibration method re-conceptualizing feature evolution as an Inertial Navigation System (INS) problem. NaviCache bridges the fundamental domain gap and the non-stationary nature of diffusion by modeling the relative coupling between input and output variations. We introduce a dual-state estimation architecture that adaptively tracks the feature change ratio and its latent drift, initialized via a specialized Initial Alignment phase. By integrating a time-dependent noise schedule with an uncertainty-aware Measurement Update mechanism, NaviCache provides a theoretically grounded mechanism for error-bounded block skipping. Extensive experiments on the HunyuanVideo, Wan, and Open-Sora series demonstrate that NaviCache exhibits more accurate error judgment for block skipping and achieves outstanding comprehensive performance.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-lv26m, title = {{N}avi{C}ache: Test-Time Self-Calibration Caching for Video Generation}, author = {Lv, Zheqi and Zhu, Zhibo and Wang, Jinke and Tian, Qi and Zhang, Shengyu and Chen, Zhengyu and Zang, Chengxi and Zhao, Zhou and Wu, Fei}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {83724--83737}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/lv26m/lv26m.pdf}, url = {https://proceedings.mlr.press/v306/lv26m.html}, abstract = {Video Diffusion Models (VDMs) is constrained by immense computational costs. While offline calibration-based acceleration suffers from calibration data dependency, prohibitive calibration duration, and susceptibility to distribution shifts, offline calibration-free methods eliminate these hurdles. However, since they rely on instantaneous zero-order approximations where the mapping between input and output differences varies in real-time, they are susceptible to observational noise and ignore the intrinsic momentum within the diffusion trajectory. In this paper, we propose NaviCache, a plug-and-play test-time self-calibration method re-conceptualizing feature evolution as an Inertial Navigation System (INS) problem. NaviCache bridges the fundamental domain gap and the non-stationary nature of diffusion by modeling the relative coupling between input and output variations. We introduce a dual-state estimation architecture that adaptively tracks the feature change ratio and its latent drift, initialized via a specialized Initial Alignment phase. By integrating a time-dependent noise schedule with an uncertainty-aware Measurement Update mechanism, NaviCache provides a theoretically grounded mechanism for error-bounded block skipping. Extensive experiments on the HunyuanVideo, Wan, and Open-Sora series demonstrate that NaviCache exhibits more accurate error judgment for block skipping and achieves outstanding comprehensive performance.} }
Endnote
%0 Conference Paper %T NaviCache: Test-Time Self-Calibration Caching for Video Generation %A Zheqi Lv %A Zhibo Zhu %A Jinke Wang %A Qi Tian %A Shengyu Zhang %A Zhengyu Chen %A Chengxi Zang %A Zhou Zhao %A Fei Wu %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-lv26m %I PMLR %P 83724--83737 %U https://proceedings.mlr.press/v306/lv26m.html %V 306 %X Video Diffusion Models (VDMs) is constrained by immense computational costs. While offline calibration-based acceleration suffers from calibration data dependency, prohibitive calibration duration, and susceptibility to distribution shifts, offline calibration-free methods eliminate these hurdles. However, since they rely on instantaneous zero-order approximations where the mapping between input and output differences varies in real-time, they are susceptible to observational noise and ignore the intrinsic momentum within the diffusion trajectory. In this paper, we propose NaviCache, a plug-and-play test-time self-calibration method re-conceptualizing feature evolution as an Inertial Navigation System (INS) problem. NaviCache bridges the fundamental domain gap and the non-stationary nature of diffusion by modeling the relative coupling between input and output variations. We introduce a dual-state estimation architecture that adaptively tracks the feature change ratio and its latent drift, initialized via a specialized Initial Alignment phase. By integrating a time-dependent noise schedule with an uncertainty-aware Measurement Update mechanism, NaviCache provides a theoretically grounded mechanism for error-bounded block skipping. Extensive experiments on the HunyuanVideo, Wan, and Open-Sora series demonstrate that NaviCache exhibits more accurate error judgment for block skipping and achieves outstanding comprehensive performance.
APA
Lv, Z., Zhu, Z., Wang, J., Tian, Q., Zhang, S., Chen, Z., Zang, C., Zhao, Z. & Wu, F.. (2026). NaviCache: Test-Time Self-Calibration Caching for Video Generation. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:83724-83737 Available from https://proceedings.mlr.press/v306/lv26m.html.

Related Material