UVCG: Leveraging Temporal Consistency for Video Protection Against Stable-Diffusion-Based Editing
IEEE Signal Processing Letters · 2026 · 33 · 2695-2699
DOI: 10.1109/lsp.2026.3705351 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Signal Processing Letters, 2026. Verfasst von Kaizhou Li und 4 weiteren Autor:innen. Die Arbeit nennt 19 Literaturangaben. Gefördert von: Shenzhen Ubiquitous Data Enabling Key Lab, Tsinghua Shenzhen International Graduate School-Shenzhen Pengrui Endowed Professorship, Shenzhen Pengrui Foundation.
Abstract (Originalsprache)
Text-guided diffusion-based video editing has become a powerful paradigm for video manipulation, but it also raises concerns about unauthorized editing. While adversarial perturbations can protect static images, extending them directly to video is unreliable because frame-wise perturbations are often weakened by temporal attention and cross-frame consistency in diffusion-based video editing pipelines. We propose Universal Video Consistency Guard (UVCG), a protection framework for diffusion-based video editing. UVCG constructs temporally coherent target signals rather than isolated frame-wise perturbations. Specifically, it introduces a distinct target video and optimizes perturbations so that the protected video follows a target-guided latent trajectory, encouraging the editor to propagate misleading features across frames instead of suppressing them as temporal jitter. We further develop a perturbation reuse strategy to exploit temporal redundancy, reduce optimization cost, and promote coherent perturbation evolution. Experiments on representative Stable Diffusion-based video editing pipelines show that UVCG reduces prompt consistency from 0.3227 to 0.2953 and achieves an 87% protection success rate in human evaluation.
Förderung
- Shenzhen Ubiquitous Data Enabling Key Lab
- Tsinghua Shenzhen International Graduate School-Shenzhen Pengrui Endowed Professorship
- Shenzhen Pengrui Foundation
Zitieren
APA
Li, K., Cao, J., He, L., Yu, X., & Zhang, X. (2026). UVCG: Leveraging Temporal Consistency for Video Protection Against Stable-Diffusion-Based Editing. IEEE Signal Processing Letters, 33, 2695-2699. https://doi.org/10.1109/lsp.2026.3705351
Vancouver
Li K, Cao J, He L, Yu X, Zhang X. UVCG: Leveraging Temporal Consistency for Video Protection Against Stable-Diffusion-Based Editing. IEEE Signal Processing Letters. 2026;33:2695-2699. doi:10.1109/lsp.2026.3705351
BibTeX
@article{Li2026,
author = {Li, Kaizhou and Cao, Junjie and He, Lang and Yu, Xinchun and Zhang, Xiao-Ping},
title = {UVCG: Leveraging Temporal Consistency for Video Protection Against Stable-Diffusion-Based Editing},
journal = {IEEE Signal Processing Letters},
year = {2026},
volume = {33},
pages = {2695--2699},
doi = {10.1109/lsp.2026.3705351}
}
EUVIMED