AuthGuard: Generalizable Deepfake Detection via Language Guidance

Abstract

Existing deepfake detection techniques struggle to keep-up with the ever-evolving novel, unseen forgeries methods. This limitation stems from their reliance on statistical artifacts learned during training, which are often tied to specific generation processes that may not be representative of samples from new, unseen deepfake generation methods encountered at test time. We propose that incorporating language guidance can improve deepfake detection generalization by integrating human-like commonsense reasoning -- such as recognizing logical inconsistencies and perceptual anomalies -- alongside statistical cues. To achieve this, we train an expert deepfake vision encoder by combining discriminative classification with image-text contrastive learning, where the text is generated by generalist MLLMs using few-shot prompting. This allows the encoder to extract both language-describable, commonsense deepfake artifacts and statistical forgery artifacts from pixel-level distributions. To further enhance robustness, we integrate data uncertainty learning into vision-language contrastive learning, mitigating noise in image-text supervision. Our expert vision encoder seamlessly interfaces with an LLM, further enabling more generalized and interpretable deepfake detection while also boosting accuracy. The resulting framework, AuthGuard, achieves state-of-the-art deepfake detection accuracy in both in-distribution and out-of-distribution settings, achieving AUC gains of 6.15% on the DFDC dataset and 16.68% on the DF40 dataset. Additionally, AuthGuard significantly enhances deepfake reasoning, improving performance by 24.69% on the DDVQA dataset.

Summary

Integrates language guidance and commonsense reasoning into deepfake detection via vision-language contrastive learning, achieving state-of-the-art generalization across unseen forgery methods.

Key Contributions

FAQ

What makes AuthGuard different from prior deepfake detectors?

AuthGuard incorporates language guidance from multimodal LLMs to learn commonsense reasoning cues (logical inconsistencies, perceptual anomalies) alongside statistical artifacts, enabling generalization to unseen forgery methods that purely statistical detectors miss.

How does the language guidance work in practice?

Generalist MLLMs generate text descriptions of deepfake artifacts via few-shot prompting. These descriptions are used in image-text contrastive learning to train a vision encoder that captures both describable and statistical forgery cues.

What is data uncertainty learning and why is it needed?

MLLM-generated text descriptions can be noisy. Data uncertainty learning models the confidence of each image-text pair, down-weighting unreliable supervision signals to make contrastive training more robust.

Does AuthGuard support interpretable detection?

Yes. The expert vision encoder interfaces with an LLM for interpretable deepfake reasoning, improving performance by 24.69% on the DDVQA reasoning benchmark.

How well does AuthGuard generalize to unseen attacks?

AuthGuard achieves state-of-the-art out-of-distribution performance, with AUC gains of 6.15% on DFDC and 16.68% on DF40, demonstrating strong generalization to novel forgery methods.

Keywords

Related Papers

• Learning Self-Consistency for Deepfake Detection (ICCV 2021)
• Principles of Designing Robust Remote Face Anti-Spoofing Systems (arXiv 2024)
• Optimal Transport-Guided Source-Free Adaptation for Face Anti-Spoofing (CVPR 2025)

BibTeX

@inproceedings{shen2025authguard,
  title={AuthGuard: Generalizable Deepfake Detection via Language Guidance},
  author={Shen, Guangyu and Li, Zhihua and Xu, Xiang and Zhao, Tianchen and Zhang, Zheng and An, Dongsheng and Tu, Zhuowen and Xing, Yifan and Zhang, Qin},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  year={2025}
}

Recommended Citation Contexts

• Vision-language alignment through CLIP-style guidance enables deepfake detectors to generalize to unseen forgery methods (Zhao et al., 2025).
• AuthGuard leverages language supervision to learn forgery-aware representations that transfer across deepfake generation pipelines (Zhao et al., 2025).