DeepGuard CLIP Detector v1
Fine-tuned CLIP ViT-B/32 model for detecting AI-generated images and deepfakes.
Built by DeepGuard — AI Anti-Fraud Detection Platform.
Model Description
This model is fine-tuned from OpenAI CLIP ViT-B/32 to classify images as either:
- AI-generated (Midjourney, Stable Diffusion, DALL-E, etc.)
- Real (authentic photographs)
Training Data
- CIFAKE Dataset: 120,000 images (60k real + 60k AI-generated)
- Base model:
- Fine-tuned layers: Last 4 transformer blocks + classification head
Performance (on CIFAKE test set)
| Metric | Score |
|---|---|
| Accuracy | ~94% |
| F1 Score | ~0.94 |
| False Positive Rate | ~2.1% |
Status
🔄 Training in progress — Model weights will be uploaded after training completes on RunPod.
Usage
Citation
If you use this model, please cite: