Venus-DefakerOne2: A Vision-Forensics Foundation Model
with Broader Coverage
GuangJian Team, Ant Group
Technical Committee on Digital Media Forensics and Security, CSIG
### 1. Environment Requirements
Python **3.12.3**
```bash
pip install torch==2.9.0 torchvision==0.24.0 ms-swift=4.1.0 numpy==1.26.0 scipy==1.11.3 scikit-learn==1.3.2 torchao==0.16.0
```
### 2. Model Download
```bash
modelscope download --model DanceText/DefakerOne2 --local_dir [target_save_path]
```
Two checkpoints are provided under the download directory:
| Directory | Description |
| --- | --- |
| `preview/` | **DefakerOne2-preview** — can be used for reproducing the scores reported in the paper. |
| `release/` | **DefakerOne2-release** — trained with more data; expected to perform better in most cases. |
### 3. Inference Data Format
Inference inputs are **JSONL** files, one sample per line.
**Image**
```json
{"messages": [{"role": "user", "content": "Is this image real or fake? Answer only Real or Fake"}], "images": ["your_image_path.jpg"]}
```
**Video**
```json
{"messages": [{"role": "user", "content": "