MediaShield — AI‑Generated & Deepfake Image Detection Platform

MediaShield is a dual‑model image forensics platform designed to detect AI‑generated images (Bing Image Creator, DALL·E, Stable Diffusion, Midjourney) and deepfake images using modern machine‑learning models. Built for reliability, clarity, and demo‑readiness, MediaShield provides a clean UI and a robust backend powered by Replicate.

---

🚀 Features

• AI‑Generated Image Detection
  Uses a diffusion‑capable Replicate model to identify images created by modern AI systems.

• Deepfake Detection
  Uses the FaceForensics++ model to detect manipulated or swapped faces.

• Dual‑Model Pipeline
  Both detectors run independently, and MediaShield combines their outputs into a final verdict.

• Clear Verdict System

  • AI‑generated
  • Deepfake
  • Real
  • Uncertain

• Clean, Professional UI
  Displays scores and verdicts with color‑coded clarity.

---

🧠 How MediaShield Works

1. Image Upload

Users upload an image through the frontend.

2. Backend Processing

The backend sends the image to two Replicate models:

Deepfake Detection

Model: FaceForensics++
Detects:

• Face swaps
• Manipulated facial regions
• Neural deepfake artifacts

Returns:
deepfake_score (0–1)

AI‑Generated Detection

Model: tstramer/ai-image-detector
Detects:

• Bing Image Creator
• DALL·E
• Stable Diffusion
• Midjourney

Returns:
ai_generated_score (0–1)

3. Verdict Logic

MediaShield combines both scores:

• AI-generated → AI score > 0.7
• Deepfake → Deepfake score > 0.7
• Uncertain → borderline scores
• Real → both scores low

4. Frontend Display

The ResultCard component shows:

• AI‑generated score
• Deepfake score
• Final verdict
• Color-coded output

---

🛠️ Tech Stack

• Next.js 14
• TypeScript
• TailwindCSS
• Replicate API
• Node.js Runtime

---

📦 Installation & Setup

1. Clone the repository

git clone https://github.com/PrithikaGopinath/MediaShield.git
cd MediaShield

2. Install dependencies

npm install

3. Add environment variables

Create .env.local:

REPLICATE_API_TOKEN=your_replicate_api_key_here

4. Run the development server

npm run dev

Your app will be live at:

http://localhost:3000

---

📡 API Endpoints

POST /api/detect

Uploads an image and returns:

{
  "ai_generated_score": 0.92,
  "deepfake_score": 0.03,
  "final_label": "AI-generated"
}

---

🧪 Testing the System

AI Images

Try images from:

• Bing Image Creator
• DALL·E
• Midjourney
• Stable Diffusion

Expected:
AI-generated Score: high

Deepfakes

Try:

• Face swaps
• Deepfake datasets

Expected:
Deepfake Score: high

Real Images

Expected:
Both scores low → Verdict: Real

---

📁 Project Structure

/components
  ResultCard.tsx

/src/app/api/detect
  route.ts

/public
  (static assets)

/app
  (frontend pages)

---

🏆 Why MediaShield Stands Out

• Detects modern diffusion AI images (most detectors fail here)
• Uses two independent forensic models
• Clean, judge‑friendly UI
• Fully reproducible setup
• Fast, reliable, and hackathon‑ready

---

🔮 Future Enhancements

• Video deepfake detection
• Metadata + EXIF analysis
• Multi‑model ensemble scoring
• Browser extension version
• Proper settign up of the model to analyse the data.