Research RadarDeepfake DetectionarXivMay 2026

Monthly arXiv Radar

May 2026 Deepfake Detection Papers: Lightweight Video Cues, Foundation-Model Limits, and Diffusion-Face Localization

Deepfake detection in May 2026 focused less on isolated benchmark wins and more on deployment pressure: smaller models, clearer generalization limits, and localization for diffusion-era face generation. These papers are useful for teams deciding whether to scale detectors, freeze foundation backbones, or invest in multimodal forensic signals.

What This Month Signals

The strongest signal this month is restraint: better deepfake defense may come from targeted cues and honest generalization audits, not only larger detectors.

Paper 012026-05-27cs.CV

Lightweight Complementary-Cue Fusion for Robust Video Face Forgery Detection

arXiv PDF

Authors & Institutions

Sunghwan Baek

Carnegie Mellon University, USA

Tariq Anwaar

Carnegie Mellon University, USA

Karanveer Singh

Carnegie Mellon University, USA

Rita Singh

Carnegie Mellon University, USA

What Problem It Solves

The paper tests whether carefully chosen handcrafted forensic cues can improve robustness without a large model.

Key Result

The extra module adds only 292 parameters, raises average AUC from 74.8% to 78.6% on FaceForensics++ and from 70.5% to 74.9% on DFDC-Preview, and outperforms F3Net, SRM, and SPSL across eight benchmarks.

Abstract

Current face video forgery detectors use wide or dual-stream backbones. We show that a single, lightweight fusion of two handcrafted cues can achieve higher accuracy with a much smaller model. Based on the Xception baseline model (21.9 million parameters), we build two detectors: LFWS, which adds a 1x1 convolution to combine a low-frequency Wavelet-Denoised Feature (WDF) with a phase-spectrum channel derived from Spatial-Phase Shallow Learning (SPSL), and LFWL, which merges WDF with Local Binary Patterns (LBP) in the same way. This extra module adds only 292 parameters, keeping the total at 21.9 million, smaller than F3Net (22.5 million) and less than half the size of SRM (55.3 million). Even with this minimal overhead, the fused models increase the average area under the curve (AUC) from 74.8% to 78.6% on FaceForensics++ and from 70.5% to 74.9% on DFDC-Preview, gains of 3.8% and 4.4% over the Xception baseline. They also consistently outperform F3Net, SRM, and SPSL in eight public benchmarks, without extra data or test-time augmentation. These results show that carefully paired, handcrafted features, combined through the lightweight fusion block, can provide competitive robustness at a significantly lower cost than comparable frequency-based detectors. Our findings suggest a need to reevaluate scale-driven design choices in face video forgery detection.

Research Starting Point

Many video face forgery detectors grow model width or add streams, which raises deployment cost.

Method

The authors add a tiny fusion block to Xception, combining Wavelet-Denoised Features with either phase-spectrum cues or Local Binary Patterns.

Paper Summary

The main lesson is that deepfake detection does not always need a larger backbone if the forensic cues are selected and fused well. By combining low-frequency wavelet-denoised features with phase or texture cues through a tiny fusion block, the paper offers a cost-conscious alternative for teams that need broader benchmark robustness without adding data, augmentation, or heavy inference cost.

Paper 022026-05-24cs.CV

Cross-Domain Generalization Limits of Vision Foundation Models in Facial Deepfake Detection

arXiv PDF

Authors & Institutions

Ibrahim Delibasoglu

Department of Software Engineering, Faculty of Computer and Information Sciences, Sakarya University, Sakarya, Türkiye

What Problem It Solves

The paper evaluates whether frozen vision foundation backbones can generalize across deepfake domains without full retraining.

Key Result

The findings show that foundation features remain discriminative for full face synthesis but localized face edits reveal fundamental limits in linear-probe evaluation structures.

Abstract

The rapid evolution of generative models has enabled the creation of hyper-realistic facial deepfakes, exposing a critical vulnerability in modern digital forensics: the inability of detectors to generalize to unseen manipulation techniques. Traditional networks suffer from representation collapse, overfitting to localized artifact fingerprints of specific training generators. This work investigates whether modern Vision Foundation Models can serve as generalizable, out-of-the-box feature extractors capable of tracking forensic anomalies across entirely unseen generative manifolds. We conduct a systematic cross-domain evaluation comparing three foundational learning paradigms: fully supervised macro-semantic features (RoPE-ViT), pure self-supervised geometric features (DINOv3), and multi-teacher agglomerative representations (NVIDIA C-RADIOv4-H). By deploying frozen backbones subjected to downstream linear probing, we map the performance limitations of these architectures on the challenging DF40 benchmark. Our empirical findings expose the intrinsic trade-offs between pre-training paradigms and parameter scale, proving that while foundation models retain high discriminative capabilities for entire face synthesis, localized face editing techniques expose fundamental boundaries in linear probe evaluation structures. Source code and model weights are available in http://github.com/mribrahim/deepfake

Research Starting Point

Foundation models are tempting as universal forensic feature extractors, but their limits under unseen face manipulations are not obvious.

Method

It compares supervised RoPE-ViT, self-supervised DINOv3, and multi-teacher NVIDIA C-RADIOv4-H using downstream linear probing on the DF40 benchmark.

Paper Summary

The paper is a useful caution against assuming that frozen vision foundation models automatically solve deepfake generalization. Its cross-domain tests show that whole-face synthesis can be easier than localized edits, so procurement and model selection should include generator-shift and manipulation-type stress tests rather than relying on average benchmark scores.

Paper 032026-05-11cs.CV

MFVLR: Multi-domain Fine-grained Vision-Language Reconstruction for Generalizable Diffusion Face Forgery Detection and Localization

arXiv PDF

Authors & Institutions

Yaning Zhang

Faculty of Computer Science and Technology, Qilu University of Technology (Shandong Academy of Sciences), Jinan, China

Tianyi Wang

School of Computing, National University of Singapore, Singapore

Zan Gao

Shandong Artificial Intelligence Institute, Qilu University of Technology (Shandong Academy of Sciences), Jinan, China

Key Laboratory of Computer Vision and System, Ministry of Education, Tianjin University of Technology, Tianjin, China

Yibo Zhao

Key Laboratory of Computer Vision and Systems, Ministry of Education, Tianjin University of Technology, Tianjin, China

Chunjie Ma

Shandong Artificial Intelligence Institute, Qilu University of Technology (Shandong Academy of Sciences), Jinan, China

Meng Wang

School of Computer Science and Information Engineering, Hefei University of Technology, Hefei, China

What Problem It Solves

The paper tackles generalizable detection and localization for diffusion-synthesized face forgeries across generators, forgery types, and datasets.

Key Result

Experiments and visualizations show state-of-the-art performance across cross-generator, cross-forgery, and cross-dataset settings.

Abstract

The swift advancement in photo-realistic face generation technology has sparked considerable concerns across society and academia, emphasizing the requirement of generalizable face forgery detection and localization methods. Prior works tend to capture face forgery patterns across multiple domains using image modality, other modalities like fine-grained texts are not comprehensively investigated, which restricts the generalization capability of models. Besides, they usually analyze facial images created by GAN, but struggle to identify and localize those synthesized by diffusion. To solve the problems, in this paper, we devise a novel multi-domain fine-grained vision-language reconstruction (MFVLR) model, which explores comprehensive and diverse visual forgery traces via language-guided face forgery representation learning, to achieve generalizable diffusion-synthesized face forgery detection and localization (DFFDL). Specifically, we devise a fine-grained language transformer that studies general fine-grained language embeddings using language reconstruction. We propose a multi-domain vision encoder to capture general and complementary visual forgery patterns across the image and residual domains. A vision decoder is designed to reconstruct image appearance and achieve forgery localization. Besides, we propose an innovative plug-and-play vision injection module to enhance the interaction between the vision and language embeddings. Extensive experiments and visualizations demonstrate that our network outperforms the state of the art on different settings like cross-generator, cross-forgery, and cross-dataset evaluations.

Research Starting Point

Diffusion face generators shift the problem from detecting old GAN artifacts to recognizing and localizing subtler, more diverse forgery traces.

Method

MFVLR combines a fine-grained language transformer, a multi-domain vision encoder over image and residual domains, a vision decoder for localization, and a plug-and-play vision injection module to strengthen vision-language interaction.

Paper Summary

MFVLR is relevant because it moves diffusion-face forensics beyond image-level yes/no detection toward localization and cross-domain explanation. By combining fine-grained language reconstruction, visual residual domains, and a decoder for forged-region localization, it can support review workflows where teams need to know not only whether an image is fake, but where the evidence appears.