---
base_model:
- timm/vit_small_patch16_384.augreg_in21k_ft_in1k
library_name: transformers
license: mit
pipeline_tag: image-classification
tags:
- image-classification
- timm
- transformers
- detection
- deepfake
- forensics
- deepfake_detection
- community
- opensight
---

# NOTICE: 7/22/2026

- Ungated again. Friendly reminder to all to please respect the MIT license. Attribution should be made to the original project (see bottom) as well as this repository. Other than that, you're free to use the model as you wish.

## Trained on 2.7M samples across 4,803 generators (see Training Data)

Model presented in [Community Forensics: Using Thousands of Generators to Train Fake Image Detectors](https://huggingface.co/papers/2411.04125).

**Uploaded for community validation as part of OpenSight** - An upcoming open-source framework for adaptive deepfake detection.

**Project OpenSight HF Spaces coming soon with an eval playground and eventually a leaderboard. Preview:** 

![image/png](/static-proxy?url=https%3A%2F%2Fcdn-uploads.huggingface.co%2Fproduction%2Fuploads%2F639daf827270667011153fbc%2FAUmW697OefKN83BClM1ae.png)

## Model Details
### Model Description
Vision Transformer (ViT) model trained on the largest dataset to-date for detecting AI-generated images in forensic applications.

- **Developed by:** Jeongsoo Park and Andrew Owens, University of Michigan
- **Model type:** Vision Transformer (ViT-Small)
- **License:** MIT (compatible with CreativeML OpenRAIL-M referenced in [2411.04125v1.pdf])
- **Finetuned from:** timm/vit_small_patch16_384.augreg_in21k_ft_in1k
- **Adapted for HF** inference compatibility by Borderless.

**HF Space will be open sourced shortly showcasing various ways to run ultra-fast inference. Make sure to follow us for updates, as we will be releasing a slew of projects in the coming weeks.**

### Links
- **Repository:** [JeongsooP/Community-Forensics](https://github.com/JeongsooP/Community-Forensics)
- **Paper:** [arXiv:2411.04125](https://arxiv.org/pdf/2411.04125)
- **Project Page:** https://jespark.net/projects/2024/community_forensics

## Training Details
### Training Data
- 2.7mil images from 15+ generators, 4600+ models
- Over 1.15TB worth of images

### Training Hyperparameters
- **Framework:** PyTorch 2.0
- **Precision:** bf16 mixed
- **Optimizer:** AdamW (lr=5e-5)
- **Epochs:** 10
- **Batch Size:** 32

## Evaluation
### Unverified Testing Results
- Only unverified because we currently lack resources to evaluate a dataset over 1.4T large.

| Metric        | Value |
|---------------|-------|
| Accuracy      | 97.2% |
| F1 Score      | 0.968 |
| AUC-ROC       | 0.992 |
| FP Rate       | 2.1%  |

![image/png](/static-proxy?url=https%3A%2F%2Fcdn-uploads.huggingface.co%2Fproduction%2Fuploads%2F639daf827270667011153fbc%2Fg-dLzxLBw1RAuiplvFCxh.png)

## Re-sampled and refined dataset

- **Coming soon™**

## Citation
**BibTeX:**
```bibtex
@misc{park2024communityforensics,
    title={Community Forensics: Using Thousands of Generators to Train Fake Image Detectors}, 
    author={Jeongsoo Park and Andrew Owens},
    year={2024},
    eprint={2411.04125},
    archivePrefix={arXiv},
    primaryClass={cs.CV},
    url={https://arxiv.org/abs/2411.04125}, 
}
```