A Generalizable and Explainable Framework for Synthetic Video Detection Using First-Digit Gradient Statistics

2026-09-30

AI Summary

This study addresses the limited interpretability and constrained cross-scenario generalization of existing AI-generated video detectors by proposing a universal detection framework grounded in the first-digit law. This work pioneers the integration of Benford’s Law into Sobel gradient statistics, combining Linear Discriminant Analysis (LDA) visualization with a Multilayer Perceptron (MLP) classifier to construct a detection mechanism fully decoupled from container formats, encoding schemes, and compression artifacts without relying on generator-specific priors. Experimental evaluations on benchmarks such as GenBuster validate the strong discriminative capability of the proposed feature set while revealing inherent limitations in zero-shot detection. Ultimately, this approach achieves AI-generated video detection that simultaneously ensures high interpretability and robust generalization across diverse scenarios.

Abstract

AI video generators have not only become harder to detect but are used to generate a diverse set of scenarios from landscapes to street views to animal videos. This creates a problem where CNN-based detectors are effective but offer no insight into their inner workings, while forensics-based detectors are often pretrained for a set scenario or become too complex to derive meaningful insights. We present a novel approach to AI video detection using Sobel gradient values analysed with the first-digit law. Using linear discriminant analysis, we visualise the discriminatory signal, while a multi-layer perceptron is used for classification. The detection method has no generator- or scenespecific features, and the model has no knowledge of container formats, codec, bitrate, or compression artefacts. The model is trained and tested on GenBuster-200K, GenBusterBench, GenVA, FaceForensics++ C23, and CelebDF. We also show how zero-shot detection fails even though the feature set carries a discriminatory signal.

Problem

• synthetic video detection
• explainability
• generalizability
• AI-generated video
• forensic detection

Innovation

• Synthetic Video Detection
• First-Digit Law
• Sobel Gradient
• Explainable AI
• Generalizable Framework