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 · 2026

Read NaviCache: Test-Time Self-Calibration Caching for Video Generation by Zheqi Lv, Zhibo Zhu, Jinke Wang, Qi Tian, Shengyu Zhang, Zhengyu Chen, Chengxi Zang, Zhou Zhao, Fei Wu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-26795.txt.

Description

Zheqi Lv and eight colleagues present NaviCache, a way to speed up video generation at run time. Video diffusion models are costly to run. Caching tricks skip repeated computation, but offline calibration needs data and time, and breaks when inputs shift. Calibration-free methods guess from instant differences, which are noisy. They also ignore the momentum of the generation trajectory.

NaviCache treats feature evolution like an inertial navigation problem. Such systems track motion by estimating drift, not by looking outside. A dual-state estimator follows the ratio of input to output change and its hidden drift. A special alignment phase starts the estimates. A time-dependent noise plan and an uncertainty-aware update step keep errors bounded while steps are skipped. The method plugs into existing models without retraining.

Catalog record

CREATORZheqi Lv, Zhibo Zhu, Jinke Wang, Qi Tian, Shengyu Zhang, Zhengyu Chen, Chengxi Zang, Zhou Zhao, Fei Wu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE47 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.26795

Browse the full library · About the Tower of Babel