DBINDS -- Can Initial Noise from Diffusion Model Inversion Help Reveal AI-Generated Videos?

TL;DR

This paper introduces DBINDS, a novel diffusion-model-inversion based detector that analyzes latent-space dynamics to distinguish real from AI-generated videos, showing strong cross-generator performance and robustness.

Contribution

The paper presents a new detection method using initial noise sequences from diffusion inversion, improving generalization over existing pixel-based detectors.

Findings

DBINDS achieves high accuracy on GenVidBench

It generalizes well across different video generators

The method is robust with limited training data

Abstract

AI-generated video has advanced rapidly and poses serious challenges to content security and forensic analysis. Existing detectors rely mainly on pixel-level visual cues and generalize poorly to unseen generators. We propose DBINDS, a diffusion-model-inversion based detector that analyzes latent-space dynamics rather than pixels. We find that initial noise sequences recovered by diffusion inversion differ systematically between real and generated videos. Building on this, DBINDS forms an Initial Noise Difference Sequence (INDS) and extracts multi-domain, multi-scale features. With feature optimization and a LightGBM classifier tuned by Bayesian search, DBINDS (trained on a single generator) achieves strong cross-generator performance on GenVidBench, demonstrating good generalization and robustness in limited-data settings.