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

arXiv: https://arxiv.org/abs/2605.24965 · HTML: https://mlsanity.com/paper/2605.24965/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 10 of its 14 references that are in the map.
- First posted on arXiv 24 May 2026.

## Builds on (most cited first, 10 of 10)

- [DINOv3](https://mlsanity.com/paper/2508.10104/) (2025)
- [StyleCLIP: Text-Driven Manipulation of StyleGAN Imagery](https://mlsanity.com/paper/2103.17249/) (2021)
- [VGGFace2: A dataset for recognising faces across pose and age](https://mlsanity.com/paper/1710.08092/) (2017)
- [Towards Universal Fake Image Detectors that Generalize Across Generative Models](https://mlsanity.com/paper/2302.10174/) (2023)
- [Rotary Position Embedding for Vision Transformer](https://mlsanity.com/paper/2403.13298/) (2024)
- [DF40: Toward Next-Generation Deepfake Detection](https://mlsanity.com/paper/2406.13495/) (2024)
- [Collaborative Diffusion for Multi-Modal Face Generation and Editing](https://mlsanity.com/paper/2304.10530/) (2023)
- [Faster Than Lies: Real-time Deepfake Detection using Binary Neural Networks](https://mlsanity.com/paper/2406.04932/) (2024)
- [C-RADIOv4 (Tech Report)](https://mlsanity.com/paper/2601.17237/) (2026)
- [Rethinking Cross-Generator Image Forgery Detection through DINOv3](https://mlsanity.com/paper/2511.22471/) (2025)

## Abstract (arXiv)

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