VidForensics-M1: Meta-Detection Reinforcement Learning with Verifiable Temporal Grounding for AI-Generated Video Forensics

2026-08-11

AI Summary

This work addresses the limitations of existing AI-generated video detection methods, which rely on coarse-grained supervision and struggle to generalize across unseen generation models and scenarios while lacking verifiable fine-grained forensic evidence. To overcome these challenges, the study introduces meta-detection into video forensics for the first time, proposing a reinforcement learning–based framework that constructs real–forged video pairs through controlled temporal segment replacement. The approach incorporates an evidence-guided reward redistribution mechanism to jointly optimize classification accuracy and temporal localization of manipulations. By integrating boundary-frame conditional generation, multimodal large language models, and an automated data pipeline, the method significantly enhances detection accuracy and generalization across diverse, previously unseen generative models and scenes, while delivering reliable and interpretable localization of forged intervals.

Abstract

Recent advances in video generation models have significantly improved the realism of synthetic videos, blurring the boundary between generated and authentic content and raising concerns about misinformation. Existing MLLM-based detectors mainly rely on supervised fine-tuning or label-level reinforcement learning, where coarse supervision limits generalization to unseen scenarios and emerging video generators. To overcome these limitations, we are the first to introduce \textbf{meta-detection} into AI-generated video detection, enabling reliable forgery detection by jointly optimizing predicted labels and supporting evidence within reinforcement learning. This paradigm requires reliable evidence signals and effective mechanisms to integrate them into label-level optimization. Textual rationales provide semantic descriptions of forgery artifacts, but their generation and verification depend on external models, making supervision vulnerable to hallucinations and semantic biases. In contrast, temporal grounding provides more objective and verifiable evidence, as manipulated intervals can be precisely controlled during forgery construction. Based on this insight, we propose an automated data construction pipeline that generates paired real-fake videos by replacing temporal segments with boundary-frame-conditioned video generation models. Furthermore, we introduce \textbf{Evidence-Guided Reward Redistribution}, which performs evidence-aware credit assignment by redistributing rewards among label-correct responses according to evidence quality. This preserves reliable label supervision while encouraging detectors to acquire fine-grained and verifiable forgery localization capabilities. Extensive experiments demonstrate that \textbf{VidForensics-M1} effectively leverages verifiable temporal evidence to achieve robust and generalizable AI-generated video detection.

Problem

• AI-generated video detection
• meta-detection
• temporal grounding
• forgery localization
• generalization

Innovation

• meta-detection
• temporal grounding
• evidence-guided reward redistribution
• AI-generated video forensics
• reinforcement learning