# LLMs Are Not Yet Ready for Deepfake Image Detection

arXiv: https://arxiv.org/abs/2506.10474 · HTML: https://mlsanity.com/paper/2506.10474/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Builds on 32 of its 54 references that are in the map.
- First posted on arXiv 12 Jun 2025.

## Builds on (most cited first, 30 of 32)

- [Visual Instruction Tuning](https://mlsanity.com/paper/2304.08485/) (2023)
- [BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models](https://mlsanity.com/paper/2301.12597/) (2023)
- [MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Models](https://mlsanity.com/paper/2304.10592/) (2023)
- [Celeb-DF: A Large-scale Challenging Dataset for DeepFake Forensics](https://mlsanity.com/paper/1909.12962/) (2019)
- [Thinking in Frequency: Face Forgery Detection by Mining Frequency-aware Clues](https://mlsanity.com/paper/2007.09355/) (2020)
- [FakeAVCeleb: A Novel Audio-Video Multimodal Deepfake Dataset](https://mlsanity.com/paper/2108.05080/) (2021)
- [Capsule-Forensics: Using Capsule Networks to Detect Forged Images and Videos](https://mlsanity.com/paper/1810.11215/) (2018)
- [Self-Supervised Video Forensics by Audio-Visual Anomaly Detection](https://mlsanity.com/paper/2301.01767/) (2023)
- [AltFreezing for More General Video Face Forgery Detection](https://mlsanity.com/paper/2307.08317/) (2023)
- [Can ChatGPT Detect DeepFakes? A Study of Using Multimodal Large Language Models for Media Forensics](https://mlsanity.com/paper/2403.14077/) (2024)
- [AntifakePrompt: Prompt-Tuned Vision-Language Models are Fake Image Detectors](https://mlsanity.com/paper/2310.17419/) (2023)
- [The GAN is dead; long live the GAN! A Modern GAN Baseline](https://mlsanity.com/paper/2501.05441/) (2025)
- [ADD: Frequency Attention and Multi-View based Knowledge Distillation to Detect Low-Quality Compressed Deepfake Images](https://mlsanity.com/paper/2112.03553/) (2021)
- [FReTAL: Generalizing Deepfake Detection using Knowledge Distillation and Representation Learning](https://mlsanity.com/paper/2105.13617/) (2021)
- [CoReD: Generalizing Fake Media Detection with Continual Representation using Distillation](https://mlsanity.com/paper/2107.02408/) (2021)
- [One Detector to Rule Them All: Towards a General Deepfake Attack Detection Framework](https://mlsanity.com/paper/2105.00187/) (2021)
- [Evaluation of an Audio-Video Multimodal Deepfake Dataset using Unimodal and Multimodal Detectors](https://mlsanity.com/paper/2109.02993/) (2021)
- [Quality-Agnostic Deepfake Detection with Intra-model Collaborative Learning](https://mlsanity.com/paper/2309.05911/) (2023)
- [Seeing is Living? Rethinking the Security of Facial Liveness Verification in the Deepfake Era](https://mlsanity.com/paper/2202.10673/) (2022)
- [A Convolutional LSTM based Residual Network for Deepfake Video Detection](https://mlsanity.com/paper/2009.07480/) (2020)
- [SoK: Systematization and Benchmarking of Deepfake Detectors in a Unified Framework](https://mlsanity.com/paper/2401.04364/) (2024)
- [Can Multi-modal (reasoning) LLMs work as deepfake detectors?](https://mlsanity.com/paper/2503.20084/) (2025)
- [Why Do Facial Deepfake Detectors Fail?](https://mlsanity.com/paper/2302.13156/) (2023)
- [A2C: A Modular Multi-stage Collaborative Decision Framework for Human-AI Teams](https://mlsanity.com/paper/2401.14432/) (2024)
- [Towards Understanding of Deepfake Videos in the Wild](https://mlsanity.com/paper/2309.01919/) (2023)
- [Exploring the Asynchronous of the Frequency Spectra of GAN-generated Facial Images](https://mlsanity.com/paper/2112.08050/) (2021)
- [Deepfake in the Metaverse: Security Implications for Virtual Gaming, Meetings, and Offices](https://mlsanity.com/paper/2303.14612/) (2023)
- [Towards Safe Synthetic Image Generation On the Web: A Multimodal Robust NSFW Defense and Million Scale Dataset](https://mlsanity.com/paper/2504.11707/) (2025)
- [UGAD: Universal Generative AI Detector utilizing Frequency Fingerprints](https://mlsanity.com/paper/2409.07913/) (2024)
- [Exploring the Impact of Moire Pattern on Deepfake Detectors](https://mlsanity.com/paper/2407.10399/) (2024)

## Built on by (most cited first, 4 of 4)

- [Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning](https://mlsanity.com/paper/2508.21048/) (2025)
- [On the Reliability of Vision-Language Models Under Adversarial Frequency-Domain Perturbations](https://mlsanity.com/paper/2507.22398/) (2025)
- [The Age of Sensorial Zero Trust: Why We Can No Longer Trust Our Senses](https://mlsanity.com/paper/2507.00907/) (2025)
- [Trustworthy Image Authentication using Forensic Knowledge Graphs](https://mlsanity.com/paper/2606.23917/) (2026)

## Abstract (arXiv)

The growing sophistication of deepfakes presents substantial challenges to the integrity of media and the preservation of public trust. Concurrently, vision-language models (VLMs), large language models enhanced with visual reasoning capabilities, have emerged as promising tools across various domains, sparking interest in their applicability to deepfake detection. This study conducts a structured zero-shot evaluation of four prominent VLMs: ChatGPT, Claude, Gemini, and Grok, focusing on three primary deepfake types: faceswap, reenactment, and synthetic generation. Leveraging a meticulously assembled benchmark comprising authentic and manipulated images from diverse sources, we evaluate each model's classification accuracy and reasoning depth. Our analysis indicates that while VLMs can produce coherent explanations and detect surface-level anomalies, they are not yet dependable as standalone detection systems. We highlight critical failure modes, such as an overemphasis on stylistic elements and vulnerability to misleading visual patterns like vintage aesthetics. Nevertheless, VLMs exhibit strengths in interpretability and contextual analysis, suggesting their potential to augment human expertise in forensic workflows. These insights imply that although general-purpose models currently lack the reliability needed for autonomous deepfake detection, they hold promise as integral components in hybrid or human-in-the-loop detection frameworks.