📜 Table of Contents
DeepScan-AI is a full-stack AI application for detecting deepfakes in videos. It leverages a custom-trained TensorFlow model to analyze individual frames and computes an overall verdict of authenticity.
| ✅ Feature | Description |
|---|---|
| 🎥 Video Upload | Drag & drop or browse to upload video |
| 📸 Frame-by-Frame Scan | Breaks video into frames and runs each through a trained model |
| 🧠 Deep Learning Model | TensorFlow CNN predicts manipulation score per frame |
| 📊 Final Verdict | Displays averaged score with verdict: ✅ Authentic / |
| 🌐 Fully Responsive | Tailwind-powered sleek and modern interface |
| 🚀 FastAPI + Express | Lightweight backend infrastructure |
DeepScan-AI is built using a modular, full-stack architecture that combines modern web technologies with machine learning. Here's a breakdown of the stack:
| Technology | Purpose |
|---|---|
| Next.js | React-based framework for frontend pages |
| React | Component-driven UI logic |
| Tailwind CSS | Utility-first CSS for styling |
| Lucide Icons | Clean, minimal icon set used in UI |
| Axios | HTTP client for interacting with the API |
| Technology | Purpose |
|---|---|
| Node.js | Runtime environment |
| Express.js | Upload route and form handling |
| Multer | Middleware for handling file uploads |
| Technology | Purpose |
|---|---|
| Python | Core language for analysis pipeline |
| FastAPI | Fast and lightweight API server |
| TensorFlow | ML framework for deepfake detection |
| Keras | High-level neural networks API (CNN model usage) |
| OpenCV | Frame extraction and image processing |
- Local Development using
localhost:3000(frontend),4000(upload), and8000(inference) - Modular Folder Structure:
/client,/server,/analyzerfor separation of concerns - REST API communication between services
💡 The stack was chosen to maximize performance, maintainability, and ease of integration between modern web and AI tools.
- Upload a video from the client.
- The Express server handles file saving.
- FastAPI backend extracts frames using OpenCV.
- Each frame is evaluated using a CNN deepfake model.
- The average of frame scores determines the authenticity verdict.
- Results are sent back to the frontend for display.
DeepScan-AI/
├── client/ # Next.js frontend
│ └── pages/
│ └── index.tsx # Main UploadPage
├── server/ # Node.js Express middleware
│ └── index.js
├── analyzer/ # Python FastAPI + TensorFlow backend
│ ├── main.py
│ ├── mesonet.py
│ └── mesonet_model.h5 # Pretrained model
├── README.md
├── .gitignoreBefore getting started with DeepScan-AI, make sure you have the following installed:
- Node.js (v18+)
- npm or yarn
- Python (3.8+)
- pip and virtualenv
Follow these steps to set up the full DeepScan-AI stack on your machine:
git clone https://github.com/your-username/DeepScan-AI
cd DeepScan-AI⚙️ 2. Set up the Deepfake Analysis Backend (FastAPI + TensorFlow)
cd analyzer
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
pip install fastapi==0.95.2 uvicorn==0.22.0 numpy==1.24.4 opencv-python==4.8.0.76 tensorflow==2.11.0 python-multipart==0.0.6 pydantic==1.10.13
uvicorn main:app --reloadThe backend will start at: http://localhost:8000
🌐 3. Start the Node.js Upload Server
cd ../server
npm install
node index.jsUpload server will run at: http://localhost:4000
🖼️ 4. Start the Next.js Frontend
cd ../client
npm install
npm run devFrontend will be live at: http://localhost:3000
- 💬 Join the Discussions: Share your insights, provide feedback, or ask questions.
- 🐛 Report Issues: Submit bugs found or log feature requests.
- 💡 Submit Pull Requests: Review open PRs, and submit your own PRs.
Built by Alexander Potiagalov