Facial video protection method prevents identity misuse in synthetic videos
KVeil: First-Place Solution to ECCV 2026 LifeGenIP Challenge on Unlearnable Videos
Computer Vision and Pattern Recognition
Summary
Creating fake videos that look like real people can be a privacy problem because just a few short clips can be used to imitate someone. The authors developed a way to subtly change original face videos that stops others from using them to make fake videos without permission. Their method works by confusing the video generation system so it can't copy the face properly, and protects both the face and background even when attackers try different tricks. This approach won first place in a challenge aiming to keep people's video identities safe.
What this means in practice
- •For video platform developers: Implement protection in user-uploaded videos to prevent unauthorized face cloning for synthetic content generation.
- •For digital privacy service providers: Offer privacy-enhancing tools that embed subtle alterations in personal videos to block AI-based identity spoofing.$Commercial implications: Enables creation of products that protect individual video data from misuse in synthetic media, appealing to privacy-conscious consumers.
Authors
Boheng Li, Yueming Xing, Yiming Li, Jie Zhang, Han Qiu, Tianwei Zhang
Abstract
The rapid development of personalized video generation has raised serious concerns over identity privacy, as a few facial videos of a person now suffice to reproduce their identity in synthetic videos. This report presents the first-place solutions to both phases of the ECCV 2026 LifeGenIP Challenge on Unlearnable Videos, which aims to protect facial videos from unauthorized customization under a limited and imperceptible perturbation budget. For Phase 1, we introduce Reference Capacity Collapse, a pair of complementary losses that degrade the information carried by the reference condition while keeping subsequent video representations dependent on it, together with a Motion-Guided EOT mechanism that improves robustness across unknown inference conditions. Phase 2 further considers tuning-based customization and unseen generative models. To address both threats within a single perturbation, we design a two-layer framework that protects the facial and background regions separately, constructing structure-aligned perturbations from a synthetic target identity in the former and strengthening the white-box disruption of Phase 1 in the latter. Our method achieves nearly perfect protection under white-box reference-based customization, and remains effective under other customization paradigms, various models, and temporal attacks. Both solutions are built upon our recent work LETHE. We hope our solutions can inspire future work on protecting facial videos against unauthorized customization. Our code is available at https://github.com/AntigoneRandy/KVeil.