Multi-Agent Forensic Reasoning for Generalizable Deepfake Video Detection

2026-08-07

AI Summary

This work addresses the limited generalization of current deepfake detection methods, which stems from the scarcity of large-scale, fine-grained annotated datasets encompassing diverse generation techniques and the inability of single models to comprehensively capture multidimensional forgery artifacts. To overcome these challenges, the authors introduce FaceVid-Forensics-100K, a dataset comprising 100,000 videos generated using 33 distinct synthesis methods, and propose the first multi-agent collaborative forensic reasoning framework. In this framework, four expert agents independently analyze forgery cues from texture, illumination, motion, and physical consistency perspectives, while a referee agent fuses their judgments to produce interpretable decisions. Conflicts are resolved and decisions aggregated through integration with a multimodal large language model. The proposed approach significantly outperforms existing methods in out-of-domain evaluations, achieving state-of-the-art performance across all metrics.

Abstract

The malicious use of generative artificial intelligence to create highly realistic deepfake videos raises serious ethical concerns and poses substantial challenges to AI safety. However, existing deepfake video benchmarks provide limited coverage of recent synthesis methods and generally lack reliable fine-grained textual annotations. Meanwhile, conventional detectors and multimodal large language models (MLLMs), whether operating as a single model or relying on a single analytical perspective, often fail to capture subtle forgery artifacts, limiting their generalization to emerging AI-generated methods. To address these limitations, we introduce FaceVid-Forensics-100K, a large-scale deepfake video dataset comprising 100,000 videos and spanning 33 synthesis methods across face swapping, face reenactment, and entire-face synthesis, including recent generators such as Seedance 2.0. The dataset provides fine-grained textual annotations of visual observations and verdict-consistent forensic explanations, automatically synthesized through a multi-model aggregation and conflict-resolution pipeline powered by advanced MLLMs. Building on this benchmark, we propose a multi-agent forensic reasoning framework that employs four specialized domain-expert agents to independently analyze forgery cues from four perspectives: texture, lighting, motion, and physics. A judge agent then reconciles their reports to produce a final prediction together with an explanation. Extensive evaluations on out-of-domain test sets show that, despite being composed entirely of small open-source MLLMs, our framework outperforms all methods including closed-source GPT and Gemini models and ranks first across all reported metrics on this benchmark. The project page is available at https://xavierjiezou.github.io/ARGUS/.

Problem

• deepfake video detection
• generalization
• forensic reasoning
• multimodal large language models
• synthesis methods

Innovation

• multi-agent reasoning
• deepfake detection
• forensic explanation
• multimodal LLMs
• generalizable AI