Organizations: University of Liverpool, UK · Hangzhou Dianzi University · Wuhan University · The Chinese University of Hong Kong, Shenzhen, Guangdong, China · Nanyang Technological University, SG
As generative video models become increasingly realistic, detecting AI-generated videos requires systems that offer both accuracy and interpretability. However, applying Multimodal Large Language Models (MLLMs) to video forensics is currently limited by outdated datasets, simplistic evaluation protocols, and a reliance on black-box classification. To address these issues, we introduce a comprehensive dataset, benchmark, and baseline model for video forgery detection. First, we present GenBuster-200K, a fair dataset of over 200,000 high-quality videos sourced from state-of-the-art generators, featuring diverse real-world scenarios. Second, we propose GenBuster-Bench, a diagnostic benchmark spanning three progressive tracks (In-Domain, Out-of-Domain, and In-the-Wild) to evaluate models across domain shifts and generational shifts. Finally, we develop BusterX, a simple MLLM baseline with Reinforcement Learning (RL) training. Instead of direct binary classification, BusterX formulates detection as a visual reasoning task, where the generated reasoning chain serves as detector itself. Experimental results demonstrate that BusterX outperforms several leading MLLMs in both detection accuracy and rationale quality.
Figures & tables
Dataset
Latest Model
Scale
Generators
High Quality
Efficient Encoding
Resolution
FPS
Fairness
Focus on Real-World
In-the-wild Data
GVD [ 5 ]
2024/01
11k
11
✗
✗
Variable
-
-
-
✓
GVF [ 30 ]
2024/03
2.8k
6
✗
✗
Variable
-
-
-
✗
GenVideo [ 9 ]
2024/04
2271k
20
✗
✗
Variable
8-24
-
-
✓
DVF [ 44 ]
2024/10
7k
8
✓
✗
Variable
8-30
-
✓
✗
GenBuster
2026/02
203k
23
✓
✓
1024x1024
24
✓
✓
✓
Table 1: Comparison with existing datasets. GenBuster outperforms prior works in scale, visual quality and fairness (balanced gender, age, and ethnicity distributions). In-the-wild refers to whether the dataset includes videos sourced from social media.
Split
Real
Fake
ID
OOD
Wild
CogVideoX
EasyAnimate
HunyuanVideo
LTX-Video
Gen3
Jimeng
Kling
Luma
Pika
Sora
Vidu
Wanx
Gen4.5
Hailuo2.3
Kling2.5 Turbo
Kling2.6
Pixverse v5
Seedance1.5 Pro
Seedance2.0
Sora2
Veo3.1
Wan2.5
Wan2.6
GenBuster-200K
100k
100k
30k
40k
10k
20k
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
GenBuster-Bench: ID
1000
1000
300
400
100
200
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
GenBuster-Bench: OOD
-
1000
-
-
-
-
100
100
100
100
100
200
150
150
-
-
-
-
-
-
-
-
-
-
-
GenBuster-Bench: Wild
-
150
-
-
-
-
-
-
-
-
-
-
-
-
10
20
15
20
10
20
15
10
10
10
10
Table 2: Data distribution for GenBuster.
Figure 1: Data construction pipeline of GenBuster.
Figure 2: Visual Examples from GenBuster-Bench.
Methods
ID (2024)
OOD (2025)
Wild (2026)
Real
Fake
ACC
Sora
Pika
Gen3
Luma
WanX
Kling
Jimeng
Vidu
Fake
ACC
Fake
ACC
Classical Detectors
3D ResNeXt [ 16 ]
84.4
60.8
72.6
31.0
75.0
39.0
42.0
48.7
22.0
65.0
46.0
44.7
64.6
39.3
61.9
Vivit [ 4 ]
92.9
64.1
78.5
74.0
52.0
69.0
53.0
62.0
55.0
33.0
68.7
60.6
76.8
46.7
69.8
VideoMAE [ 49 ]
93.3
64.9
79.1
48.5
79.0
65.0
52.0
76.7
24.0
55.0
72.0
59.5
76.4
50.7
72.0
DeMamba [ 9 ]
93.8
70.2
82.0
65.0
75.0
73.0
56.0
46.0
68.0
71.0
84.7
66.9
80.4
48.0
70.9
Table 3: Comparison of state-of-the-art methods on GenBuster-Bench. The benchmark comprises three progressive tracks: ID evaluates performance on seen generators; OOD assesses cross-generator generalization on unseen commercial models; and Wild stress-tests real-world robustness.
Model
Score
GPT-5.2 (High) [ 34 ]
48.9
Grok-4.1-fast [ 53 ]
81.0
Claude-Sonnet-4.6 [ 3 ]
78.3
BusterX
84.7
Table 4: Explanation scores on correctly predicted Wild samples.
Figure 3: Case Study. Faced with highly realistic synthetic video, MLLMs like GPT-5.2 and Claude-Sonnet-4.6 fail, whereas BusterX successfully performs frame-level spatiotemporal localization to uncover underlying physical inconsistencies. Full responses are provided in Appendix G .
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 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 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 VidForensics-M1 effectively leverages verifiable temporal evidence to achieve robust and generalizable AI-generated video detection.
Bowei Liu, Zheng Lu, Yuhan Bian +8
Tsinghua University · Peking University · Renmin University of China +1
The rapid advancement of AI-generated video poses challenges to digital authenticity and security. Current detection methods, often trained on specific datasets, struggle with the ever-evolving landscape of generative techniques and unseen manipulations. We introduce a framework leveraging Vision Language Models (VLMs) for robust AI-generated video detection. Our approach equips the VLM with the ability to reason about video content and use external tools to identify subtle inconsistencies, mirroring human system 2 thinking. Our self-evolving VLM dynamically selects and composes appropriate tools, enhancing its ability to generalize to novel video generation techniques. The modular design promotes interpretability, allowing for a clearer understanding of VLM's decision-making process. To evaluate, we establish the first benchmark VidForensic containing 1.4k+ high-quality AI-generated videos across eight generative models. Experiments show that REVEAL improves F1 scores by 9.1% to 30.2% over top baselines across our datasets for VLMs, notably for GPT-4o, Gemini 1.5 pro, and QWen-VL-Max, and Llava-One-Vision-7B. While open-world AI-video detection remains an open challenge, our results indicate that existing methods fail primarily because they lack tool-enabled, higher-order reasoning.
Rapid advances in AI-generated video (AIGV) have increased the risks posed by deceptive video manipulation. Unlike fully synthetic videos, manipulated videos retain most source content and alter only localized regions, making forensic analysis particularly challenging. Existing video forgery research faces two limitations in both data and methodology: (1) High-quality datasets and benchmarks tailored for manipulated videos remain scarce. (2) Multimodal large language models (MLLMs) extend forgery analysis beyond binary classification but struggle to use low-level forensic cues and provide precise pixel-level grounding. Specifically, we introduce ManiVid, a unified forensic analysis task covering forgery detection, artifact grounding, and anomaly explanation for manipulated videos. We construct ManiVid-38K, the first dataset to combine paired, open-vocabulary localized manipulations of general videos with authenticity labels, forgery masks, and anomaly explanations. It comprises about 19K manually verified real-fake video pairs, mostly at 1080P resolution, generated under 2 paradigms with 15 powerful generation models. We sample 1K pairs for ManiVidBench, balanced across six manipulation types and generation models for fair evaluation. We further propose ManiVidLens, a unified framework for explainable video forgery analysis. Its Forensic Evidence Router supplies shared low-level forensic evidence for multimodal reasoning and video segmentation. Its Prompt Distill Module converts grounding states into semantic and geometric prompts and distills spatial priors for mask decoding and full-video propagation. ManiVidLens achieves relative gains over the strongest comparison methods in artifact grounding (+21.1% mIoU; +21.3% J&F) and anomaly explanation (+131.3% ROUGE-L; +9.9% CSS). Its forgery detection remains comparable to dedicated classifiers (0.914 Acc; 0.913 F1).
Hengrui Kang, Zhonghao Yan, Yuxuan Yang +7
Shanghai Jiao Tong University · Shanghai AI Laboratory · Beijing University of Posts and Telecommunications +3