# VRAG-DFD: Verifiable Retrieval-Augmentation for MLLM-based Deepfake Detection Source: [VRAG-DFD: Verifiable Retrieval-Augmentation for MLLM-based Deepfake Detection]() Collection index: [llms.txt]() Data updated: 2026-04-21T05:00:04.784Z Authors: Hui Han, Shunli Wang, Yandan Zhao, Taiping Yao, Shouhong Ding Published: 2026-04-15T09:27:49.000Z ## Abstract In Deepfake Detection (DFD) tasks, researchers proposed two types of MLLM-based methods: complementary combination with small DFD detectors, or static forgery knowledge injection. The lack of professional forgery knowledge hinders the performance of these DFD-MLLMs. To solve this, we deeply considered two insightful issues: How to provide high-quality associated forgery knowledge for MLLMs? AND How to endow MLLMs with critical reasoning abilities given noisy reference information? Notably, we attempted to address above two questions with preliminary answers by leveraging the combination of Retrieval-Augmented Generation (RAG) and Reinforcement Learning (RL). Through RAG and RL techniques, we propose the VRAG-DFD framework with accurate dynamic forgery knowledge retrieval and powerful critical reasoning capabilities. Specifically, in terms of data, we constructed two datasets with RAG: Forensic Knowledge Database (FKD) for DFD knowledge annotation, and Forensic Chain-of-Thought Dataset (F-CoT), for critical CoT construction. In terms of model training, we adopt a three-stage training method (Alignment-\>SFT-\>GRPO) to gradually cultivate the critical reasoning ability of the MLLM. In terms of performance, VRAG-DFD achieved SOTA and competitive performance on DFD generalization testing. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Face Forgery Detection on Celeb-DF v2](): Face Forgery Detection; Celeb-DF v2; Video-level AUC: 95.97. - [Deepfake Detection on WildDeepfake (WDF)](): Deepfake Detection; WildDeepfake (WDF); Video-level AUC: 0.8896. - [Face Forgery Detection on Celeb-DF v1](): Face Forgery Detection; Celeb-DF v1; Video-level AUC: 99.6. - [Face Forgery Detection on Face Forensics in the Wild (FFIW)](): Face Forgery Detection; Face Forensics in the Wild (FFIW); Video-level AUC: 93.49. - [Explanation Quality Evaluation on DFD 100 randomly selected samples (test)](): Explanation quality evaluation; DFD 100 randomly selected samples (test); GPT-4o Score: 7.55. - [Face Forgery Detection on DFDC](): Face Forgery Detection; DFDC; AUC: not recorded. ## Original sources and implementation [Original paper]() [PDF]()