The Alpha Blending Hypothesis: Compositing Shortcut in Deepfake Detection

TL;DR

This paper introduces the Alpha Blending Hypothesis, suggesting deepfake detectors mainly search for compositing artifacts rather than semantic anomalies, and proposes a new method achieving state-of-the-art generalization.

Contribution

The paper formulates the Alpha Blending Hypothesis, validates it experimentally, and develops BlenD, a method that improves cross-dataset deepfake detection without using generated deepfakes during training.

Findings

Deepfake detectors are highly sensitive to self-blended images and non-generative manipulations.

BlenD achieves the best average cross-dataset generalization on 15 datasets.

Ensemble of blending searchers and resilient models reaches 94.0% AUROC.

Abstract

Recent deepfake detection methods demonstrate improved cross-dataset generalization, yet the underlying mechanisms remain underexplored. We introduce the Alpha Blending Hypothesis, positing that state-of-the-art frame-based detectors primarily function as alpha blending searchers; rather than learning semantic anomalies or specific generative neural fingerprints, they localize low-level compositing artifacts introduced during the integration of manipulated faces into target frames. We experimentally validate the hypothesis, demonstrating that deepfake detectors exhibit high sensitivity to the so-called self-blended images (SBI) and non-generative manipulations. We propose the method BlenD that leverages a large-scale, diverse dataset of real-only facial images augmented with SBI. This approach achieves the best average cross-dataset generalization on 15 compositional deepfake datasets released between 2019 and 2025 without utilizing explicitly generated deepfakes during training. Furthermore, we show that predictions from explicit blending searchers and models resilient to blending shortcuts are highly complementary, yielding a state-of-the-art AUROC of 94.0% in an ensemble configuration. The code with experiments and the trained model will be publicly released.