1Shanghai Artificial Intelligence Laboratory, 2Sun Yat-sen University 3Beihang University, 4Shanghai Jiao Tong University, 5The Chinese University of Hong Kong, Shenzhen
[2025.9.24]: 🎉 FakeVLM was accepted to NeurIPS 2025!
[2025.4.15]: 🤗 We are excited to release the FakeClue dataset. Check out here.
[2025.3.20]: 🔥 We have released Spot the Fake: Large Multimodal Model-Based Synthetic Image Detection with Artifact Explanation. Check out the paper. We present FakeClue dataset and FakeVLM model.
FakeVLM Overview
With the rapid advancement of Artificial Intelligence Generated Content (AIGC) technologies, synthetic images have become increasingly prevalent in everyday life, posing new challenges for authenticity assessment and detection. Despite the effectiveness of existing methods in evaluating image authenticity and locating forgeries, these approaches often lack human interpretability and do not fully address the growing complexity of synthetic data. To tackle these challenges, we introduce FakeVLM, a specialized large multimodal model designed for both general synthetic image and DeepFake detection tasks. FakeVLM not only excels in distinguishing real from fake images but also provides clear, natural language explanations for image artifacts, enhancing interpretability. Additionally, we present FakeClue, a comprehensive dataset containing over 100,000 images across seven categories, annotated with fine-grained artifact clues in natural language. FakeVLM demonstrates performance comparable to expert models while eliminating the need for additional classifiers, making it a robust solution for synthetic data detection. Extensive evaluations across multiple datasets confirm the superiority of FakeVLM in both authenticity classification and artifact explanation tasks, setting a new benchmark for synthetic image detection.
framework
Contributions
We propose FakeVLM, a multimodal large model designed for both general synthetic and deepfake image detection tasks. It excels at distinguishing real from fake images while also providing excellent interpretability for artifact details in synthetic images.
We introduce the FakeClue dataset, which includes a rich variety of image categories and fine-grained artifact annotations in natural language.
Our method has been extensively evaluated on multiple datasets, achieving outstanding performance in both synthetic detection and abnormal artifact explanation tasks.
🛠️ Installation
Please clone our repository and change to that folder
Replace data paths with yours in scripts/train.sh and the original llava-1.5-7b-hf model path with yours in supported_models.py.
bash train.sh
3. Evaluation
We prepared two scripts for you to evaluate the FakeVLM model. The trained FakeVLM model is available at here.
1. Usual evaluation
bash scripts/eval.sh
2. Evaluation with vllm
Considering the size of the model and the magnitude of the data, we recommend using vllm for evaluation. Please make sure that you have installed vllm.
# change scripts/eval.py to scripts/eval_vllm.py in scripts/eval.sh
bash scripts/eval.sh
📊 Results
Performance of 7 leading LMMs and FakeVLM on DD-VQA, Fake Clues and Loki.
FakeClue
Ours dataset.
LOKI
A new benchmark for evaluating multimodal models in synthetic detection tasks. It includes human-annotated fine-grained image artifacts, enabling deeper analysis of artifact explanations. We used its image modality, covering categories like Animals, Humans, Scenery, and Documents.
framework
DD-VQA
A dataset for explaining facial artifacts, using manual annotations in a VQA format. Artifacts include blurred hairlines, mismatched eyebrows, rigid pupils, and unnatural shadows. It builds on FF++ data and emphasizes common-sense reasoning.
framework
To provide a comprehensive comparison of the model performance across the three datasets—FakeClue, LOKI, and DD-VQA—we present the following radar chart. This chart visually highlights the strengths and weaknesses of the 7 leading LMMs and FakeVLM, offering a clear depiction of their results in synthetic detection and artifact explanation tasks.
result
😄 Acknowledgement
This repository is built upon the work of LLaVA, and our codebase is built upon lmms-finetune. We appreciate their contributions and insights that have provided a strong foundation for our research.
📨 Contact
If you have any questions or suggestions, please feel free to contact us
at 466439420gh@gmail.com.
📝 Citation
If you find our work interesting and helpful, please consider giving our repo a star. Additionally, if you would like to cite our work, please use the following format:
shell
1@article{wen2025spot,
2title={Spot the fake: Large multimodal model-based synthetic image detection with artifact explanation},
3author={Wen, Siwei and Ye, Junyan and Feng, Peilin and Kang, Hengrui and Wen, Zichen and Chen, Yize and Wu, Jiang and Wu, Wenjun and He, Conghui and Li, Weijia},
4journal={arXiv preprint arXiv:2503.14905},
5year={2025}6}
