# The Alpha Blending Hypothesis: Compositing Shortcut in Deepfake Detection

arXiv: https://arxiv.org/abs/2605.10334 · HTML: https://mlsanity.com/paper/2605.10334/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 46 of its 60 references that are in the map.
- First posted on arXiv 11 May 2026.

## Builds on (most cited first, 30 of 46)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [DINOv3](https://mlsanity.com/paper/2508.10104/) (2025)
- [FaceForensics++: Learning to Detect Manipulated Facial Images](https://mlsanity.com/paper/1901.08971/) (2019)
- [Cyclical Learning Rates for Training Neural Networks](https://mlsanity.com/paper/1506.01186/) (2015)
- [Celeb-DF: A Large-scale Challenging Dataset for DeepFake Forensics](https://mlsanity.com/paper/1909.12962/) (2019)
- [Towards Universal Fake Image Detectors that Generalize Across Generative Models](https://mlsanity.com/paper/2302.10174/) (2023)
- [Detecting Deepfakes with Self-Blended Images](https://mlsanity.com/paper/2204.08376/) (2022)
- [Perception Encoder: The best visual embeddings are not at the output of the network](https://mlsanity.com/paper/2504.13181/) (2025)
- [Exposing Deep Fakes Using Inconsistent Head Poses](https://mlsanity.com/paper/1811.00661/) (2018)
- [Detecting and Simulating Artifacts in GAN Fake Images](https://mlsanity.com/paper/1907.06515/) (2019)
- [Lips Don't Lie: A Generalisable and Robust Approach to Face Forgery Detection](https://mlsanity.com/paper/2012.07657/) (2020)
- [Generalizing Face Forgery Detection with High-frequency Features](https://mlsanity.com/paper/2103.12376/) (2021)
- [LivePortrait: Efficient Portrait Animation with Stitching and Retargeting Control](https://mlsanity.com/paper/2407.03168/) (2024)
- [FakeAVCeleb: A Novel Audio-Video Multimodal Deepfake Dataset](https://mlsanity.com/paper/2108.05080/) (2021)
- [UCF: Uncovering Common Features for Generalizable Deepfake Detection](https://mlsanity.com/paper/2304.13949/) (2023)
- [DeepfakeBench: A Comprehensive Benchmark of Deepfake Detection](https://mlsanity.com/paper/2307.01426/) (2023)
- [Leveraging Real Talking Faces via Self-Supervision for Robust Forgery Detection](https://mlsanity.com/paper/2201.07131/) (2022)
- [Transcending Forgery Specificity with Latent Space Augmentation for Generalizable Deepfake Detection](https://mlsanity.com/paper/2311.11278/) (2023)
- [Orthogonal Subspace Decomposition for Generalizable AI-Generated Image Detection](https://mlsanity.com/paper/2411.15633/) (2024)
- [Face Forensics in the Wild](https://mlsanity.com/paper/2103.16076/) (2021)
- [KoDF: A Large-scale Korean DeepFake Detection Dataset](https://mlsanity.com/paper/2103.10094/) (2021)
- [Exposing GAN-generated Faces Using Inconsistent Corneal Specular Highlights](https://mlsanity.com/paper/2009.11924/) (2020)
- [LatentSync: Taming Audio-Conditioned Latent Diffusion Models for Lip Sync with SyncNet Supervision](https://mlsanity.com/paper/2412.09262/) (2024)
- [In Ictu Oculi: Exposing AI Generated Fake Face Videos by Detecting Eye Blinking](https://mlsanity.com/paper/1806.02877/) (2018)
- [Face Forgery Detection by 3D Decomposition](https://mlsanity.com/paper/2011.09737/) (2020)
- [SeeABLE: Soft Discrepancies and Bounded Contrastive Learning for Exposing Deepfakes](https://mlsanity.com/paper/2211.11296/) (2022)
- [Diff2Lip: Audio Conditioned Diffusion Models for Lip-Synchronization](https://mlsanity.com/paper/2308.09716/) (2023)
- [FreqBlender: Enhancing DeepFake Detection by Blending Frequency Knowledge](https://mlsanity.com/paper/2404.13872/) (2024)
- [MEMO: Memory-Guided Diffusion for Expressive Talking Video Generation](https://mlsanity.com/paper/2412.04448/) (2024)
- [Lips Are Lying: Spotting the Temporal Inconsistency between Audio and Visual in Lip-Syncing DeepFakes](https://mlsanity.com/paper/2401.15668/) (2024)

## Abstract (arXiv)

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.