G2VD: Generalizable AI-Generated Video Detection via Counterfactual Intervention and Causal Disentanglement

2026-07-05

AI Summary

This work addresses the limited cross-domain generalization of existing AI-generated video detection methods, which often rely on generator-specific artifacts. The authors propose G2VD, the first framework to integrate counterfactual intervention and causal disentanglement into this task. It leverages a variational autoencoder to generate controlled counterfactual samples and aligns representations in both frequency and pixel domains to steer the model toward generator-agnostic forgery signatures. A dual-branch causal disentanglement classifier is further introduced, employing Hilbert–Schmidt Independence Criterion (HSIC) constraints to separate task-relevant features from domain-specific biases. This approach substantially mitigates shortcut learning, achieving strong cross-domain performance with only 10% of training data. On four public benchmarks, G2VD significantly outperforms existing baselines, attaining over 90% accuracy and an AUC of 0.95 under the challenging GenVidBench setting.

Abstract

The rapid advancement of AI-generated videos poses increasing security risks and calls for robust detectors with strong cross-domain generalization. Although existing methods achieve promising results under in-domain evaluation, their performance often degrades substantially when tested on unseen generators. A key reason is shortcut learning, where detectors rely on domain-specific spurious cues, such as generator-dependent fingerprints and generation styles, instead of intrinsic forgery traces. To address this issue, we propose G2VD, a Generalizable AI-Generated Video Detection framework based on counterfactual intervention and causal disentanglement. First, G2VD introduces a counterfactual intervention pipeline (CFIPipeline) that generates controlled counterfactual samples via variational autoencoders (VAEs), followed by frequency-domain and pixel-domain alignment, thereby encouraging the detector to focus on generator-intrinsic cues. Building on this intervention process, we further design a causal disentanglement classifier consisting of two domain-anchored branches with distinct classification objectives, combined with an HSIC-based independence constraint to encourage the separation of task-relevant cues from domain-specific bias. Across four public datasets, G2VD shows strong average cross-domain performance and consistent gains over matched backbones. On the challenging GenVidBench cross-domain setting, it exceeds 90% accuracy and reaches an AUC close to 0.95. Notably, this performance is obtained using only 10% of the original training data. The code is available at https://github.com/dumeng98/G2VD.

Problem

• AI-generated video detection
• cross-domain generalization
• shortcut learning
• spurious cues
• forgery traces

Innovation

• counterfactual intervention
• causal disentanglement
• cross-domain generalization
• AI-generated video detection
• spurious correlation