ML Security PapersGrab-3D: Detecting AI-Generated Videos from 3D Geometric Temporal Consistency
Wenhan Chen , Sezer Karaoglu , Theo Gevers
Published on arXiv
2512.13665
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
Grab-3D significantly outperforms state-of-the-art detectors with robust cross-domain generalization to unseen video generators.
Grab-3D
Novel technique introduced
Recent advances in diffusion-based generation techniques enable AI models to produce highly realistic videos, heightening the need for reliable detection mechanisms. However, existing detection methods provide only limited exploration of the 3D geometric patterns present in generated videos. In this paper, we use vanishing points as an explicit representation of 3D geometry patterns, revealing fundamental discrepancies in geometric consistency between real and AI-generated videos. We introduce Grab-3D, a geometry-aware transformer framework for detecting AI-generated videos based on 3D geometric temporal consistency. To enable reliable evaluation, we construct an AI-generated video dataset of static scenes, allowing stable 3D geometric feature extraction. We propose a geometry-aware transformer equipped with geometric positional encoding, temporal-geometric attention, and an EMA-based geometric classifier head to explicitly inject 3D geometric awareness into temporal modeling. Experiments demonstrate that Grab-3D significantly outperforms state-of-the-art detectors, achieving robust cross-domain generalization to unseen generators.
Key Contributions
- Identifies 3D geometric temporal consistency (via vanishing points) as a discriminative forensic signal distinguishing real from AI-generated videos
- Introduces Grab-3D, a geometry-aware transformer with geometric positional encoding, temporal-geometric attention, and an EMA-based classifier head
- Constructs a static-scene AI-generated video dataset enabling stable 3D geometric feature extraction and cross-domain generalization benchmarking
🛡️ Threat Analysis
Proposes a novel AI-generated video detection framework — detecting AI-generated content (videos produced by diffusion models) is explicitly within ML09's scope of output integrity and content provenance. The novelty lies in the forensic insight (vanishing-point-based 3D geometric inconsistency) and the detection architecture, not merely applying existing detectors to a new domain.