Wasn’t Me: Enabling Users to Falsify Deepfake Attacks

Tal Reiss, Bar Cavia, Yedid Hoshen

Keywords: deepfake detection, deepfake verification

Abstract:

The rise of deepfake technology has made everyone vulnerable to false claims based on manipulated media. While many existing deepfake detection methods aim to identify fake media, they often struggle with deepfakes created by new generative models not seen during training. In this paper, we propose VeriFake, a method that enables users to verify that media claiming to show them are false. VeriFake is based on two key assumptions: (i) generative models struggle to exactly depict a specific identity, and (ii) they often fail to perfectly synchronize generated lip movements with speech. By combining these assumptions with powerful modern representation encoders, VeriFake achieves highly effective results, even against previously unseen deepfakes. Through extensive experiments, we demonstrate that VeriFake significantly outperforms general-purpose deepfake detection techniques despite being simple to implement and not relying on any fake data for pretraining.

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 3523