README.md

metadata

title: AI Image Detection
emoji: 🔍
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false
license: mit
models:
  - nahid112376/ai-detection-cnn

🔍 AI Image Detection

Upload an image to detect if it was generated by AI or is a real photograph.

How It Works

1. Feature Extraction: Images are processed using Qwen2.5-VL-3B-Instruct to extract spatial features
2. CNN Classification: A 1D CNN trained on these features detects AI generation patterns
3. Prediction: Returns a confidence score and classification (AI-generated vs Real)

Model Details

• Architecture: 1D CNN with 3 convolutional layers
• Feature Extractor: Qwen2.5-VL-3B-Instruct
• Framework: PyTorch
• Model Repository: nahid112376/ai-detection-cnn

Usage

1. Upload an image using the upload button, clipboard, or webcam
2. Adjust the detection threshold (default: 0.5)
   • Higher threshold = stricter AI detection
   • Lower threshold = more sensitive detection
3. Click "Detect AI" to get results

Technical Details

• Input: Images of any size (automatically resized)
• Output: Binary classification with confidence score
• Threshold: Adjustable from 0.0 to 1.0
• Processing: First prediction may take longer due to model loading

Performance

The model was trained on the demectai dataset and uses spatial features from Qwen2.5-VL for robust AI image detection.

Citation

@misc{ai-detection-cnn,
  author = {bartazable},
  title = {AI Detection CNN Model},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/nahid112376/ai-detection-cnn}}
}

License

MIT License