# Layer 4: Language-Guided World-Level Reasoning [Back to methods](README.md) · [Back to home](../README.md) World-grounded verification. The detector moves beyond within-video agreement and asks whether the implied claims remain consistent with external facts, commonsense, physical rules, and explicit evidence chains. ## Subcategory Guide - **A. Prompts and adapters for representation calibration**: Language priors and lightweight tuning to bind forgery evidence to semantic hypotheses. - **B. Tool-augmented agents for evidence gathering**: Observe-tool-integrate pipelines, retrieval, and modular analyzers. - **C. Post-training, preferences and rewards**: Post-training for explainable reasoning, preference alignment, and reward modeling. ## A. Prompts and adapters for representation calibration Language priors and lightweight tuning to bind forgery evidence to semantic hypotheses. | Date | Method | Paper | Base model / system | Training | Output | Venue | | --- | --- | --- | --- | --- | --- | --- | | 07/2025 | LVLMDFD | [Unlocking the Capabilities of Large Vision-Language Models for Generalizable and Explainable Deepfake Detection](https://proceedings.mlr.press/v267/yu25d.html)
Yu et al. | LVLM + LLM | Detector+prompt learner | Verdict+loc.+exp. | ICML | | 06/2025 | AuthGuard | [AuthGuard: Generalizable Deepfake Detection via Language Guidance](https://arxiv.org/abs/2506.04501)
Shen et al. | Vision encoder + LLM | Cls+ITC; uncertainty | Verdict+exp. | arXiv | | 04/2025 | RepDFD | [Standing on the Shoulders of Giants: Reprogramming Visual-Language Model for General Deepfake Detection](https://ojs.aaai.org/index.php/AAAI/article/view/32559)
Lin et al. | Frozen VLM | Prompt tuning | Label | AAAI | | 01/2025 | DeepFake-Adapter | [DeepFake-Adapter: Dual-Level Adapter for DeepFake Detection](http://dx.doi.org/10.1007/s11263-024-02274-6)
Shao et al. | ViT (frozen) | Adapter tuning | Frame label (agg.) | IJCV | | 11/2024 | CPML | [Prompt-guided Multi-modal contrastive learning for Cross-compression-rate Deepfake Detection](https://papers.bmvc2024.org/0619.pdf)
Lai et al.

[On Using rPPG Signals for DeepFake Detection: A Cautionary Note](https://doi.org/10.1007/978-3-031-43153-1_20)
D'Amelio et al. | rPPG+landmark encoder | Prompt-guided | Score | BMVC
ICIAP | | 06/2024 | ChatGPT Detect | [Can ChatGPT Detect DeepFakes? A Study of Using Multimodal Large Language Models for Media Forensics](https://openaccess.thecvf.com/content/CVPR2024W/WMF/papers/Jia_Can_ChatGPT_Detect_DeepFakes_A_Study_of_Using_Multimodal_Large_CVPRW_2024_paper.pdf)
Jia et al.

[How Good is ChatGPT at Audiovisual Deepfake Detection: A Comparative Study of ChatGPT, AI Models and Human Perception](https://www.nowpublishers.com/article/OpenAccessDownload/SIP-20250004)
Shahzad et al. | ChatGPT | Prompt | Verdict+exp. | CVPR
APSIPA Transactions on Signal and Information Processing | ## B. Tool-augmented agents for evidence gathering Observe-tool-integrate pipelines, retrieval, and modular analyzers. | Date | Method | Paper | Base model / system | Training | Output | Venue | | --- | --- | --- | --- | --- | --- | --- | | 12/2025 | DeepAgent | [DeepAgent: A Dual Stream Multi Agent Fusion for Robust Multimodal Deepfake Detection](https://arxiv.org/abs/2512.07351)
Zaman et al. | Multi-agent pipeline | Not stated | MM verdict | arXiv | | 08/2025 | FakeHunter | [Memory-Anchored Multimodal Reasoning for Explainable Video Forensics](https://arxiv.org/abs/2508.14581)
Chen et al. | Vision+audio encoders + VLM agent | Agent+retrieval | Verdict+evidence | arXiv | | 06/2025 | DAVID-XR1 | [DAVID-XR1: Detecting AI-Generated Videos with Explainable Reasoning](https://arxiv.org/abs/2506.14827)
Gao et al. | Video VLM | SFT | Loc.+defect reasoning | arXiv | | 02/2025 | LAVID | [LAVID: An Agentic LVLM Framework for Diffusion-Generated Video Detection](https://arxiv.org/abs/2502.14994)
Liu et al. | LVLM agent + tools | Tool loop | Verdict+evidence | arXiv | ## C. Post-training, preferences and rewards Post-training for explainable reasoning, preference alignment, and reward modeling. | Date | Method | Paper | Base model / system | Training | Output | Venue | | --- | --- | --- | --- | --- | --- | --- | | 02/2026 | VideoVeritas | [VideoVeritas: AI-Generated Video Detection via Perception Pretext Reinforcement Learning](https://doi.org/10.48550/ARXIV.2602.08828)
Tan et al. | Video MLLM | Pref. align + PPRL | Verdict+reasoning | arXiv | | 12/2025 | Skyra and Skyra-RL | [Skyra: AI-Generated Video Detection via Grounded Artifact Reasoning](https://arxiv.org/abs/2512.15693)
Li et al. | Qwen2.5-VL | SFT -> RL | Verdict+grounded reasoning | arXiv | | 10/2025 | VidGuard-R1 | [VidGuard-R1: AI-Generated Video Detection and Explanation via Reasoning MLLMs and RL](https://arxiv.org/abs/2510.02282)
Park et al. | Qwen2.5-VL | SFT -> DPO -> GRPO | Verdict+exp. | arXiv | | 10/2025 | EDVD-LLaMA | [EDVD-LLaMA: Explainable Deepfake Video Detection via Multimodal Large Language Model Reasoning](https://arxiv.org/abs/2510.16442)
Sun et al. | Qwen2.5-7B + video encoder | SFT | Verdict+exp. | arXiv | | 09/2025 | DeeptraceReward | [Learning Human-Perceived Fakeness in AI-Generated Videos via Multimodal LLMs](https://arxiv.org/abs/2509.22646)
Fu et al. | VideoLLaMA3 and Qwen2.5-VL with reward modeling | Reward-model train | Reward | arXiv | | 08/2025 | VERITAS | [Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning](https://arxiv.org/abs/2508.21048)
Tan et al. | InternVL3 and Qwen2.5-VL | SFT -> MiPO -> P-GRPO | Verdict+exp. | arXiv | | 07/2025 | BusterX++ | [BusterX++: Towards Unified Cross-Modal AI-Generated Content Detection and Explanation with MLLM](https://arxiv.org/abs/2507.14632)
Wen et al. | Qwen2.5-VL | RL -> SFT -> RL | Verdict+struct. exp. | arXiv | | 05/2025 | BusterX | [BusterX: MLLM-Powered AI-Generated Video Forgery Detection and Explanation](https://arxiv.org/abs/2505.12620)
Wen et al. | Qwen2.5-VL | SFT -> RL | Verdict+exp. | arXiv | | 10/2024 | X2-DFD | [X2-DFD: A framework for eXplainable and eXtendable Deepfake Detection](https://arxiv.org/abs/2410.06126)
Chen et al. | LLaVA + aux detectors | SFT (explainable data) | Verdict+exp. | arXiv |
Additional cited works discussed in this layer