██████╗ █████╗ ████████╗███████╗██╗███████╗██╗  ██╗
██╔════╝██╔══██╗╚══██╔══╝██╔════╝██║██╔════╝██║  ██║
██║     ███████║   ██║   █████╗  ██║███████╗███████║
██║     ██╔══██║   ██║   ██╔══╝  ██║╚════██║██╔══██║
╚██████╗██║  ██║   ██║   ██║     ██║███████║██║  ██║
 ╚═════╝╚═╝  ╚═╝   ╚═╝   ╚═╝     ╚═╝╚══════╝╚═╝  ╚═╝
                D E T E C T O R

Upload a photo. Get the truth.

Reverse image search + AI face analysis + metadata forensics — combined into a single risk score.

Features · Demo · How It Works · Setup · Tech Stack

---

Demo

Drop Zone

Image Selected

Analysis Result — 75/100 High Risk

Photo found on 12 different sites including Wikimedia Commons, X, Reddit, and Instagram. Face analysis detected a centered male face, age ~40, with professional-quality sharpness. EXIF metadata present but missing camera and software info.

---

Features

• Reverse Image Search — Searches Google Lens, Yandex, and SerpAPI simultaneously to find where a photo has been used online. Filters out shopping sites to only surface real matches.
• AI Face Analysis — Detects faces using OpenCV, estimates age/gender/emotion via DNN models, and flags professional photo indicators (studio lighting, high symmetry, centered composition).
• Metadata Forensics — Extracts EXIF data to detect stripped metadata, editing software traces, missing camera info, and GPS coordinates.
• Weighted Risk Score — Combines all three signals (50% reverse image, 30% face, 20% metadata) into a 0-100 catfish probability score with a clear verdict.
• Dark Mode Dashboard — Minimalist UI with drag-and-drop upload, animated score gauge, and expandable analysis breakdown.

How It Works

                    ┌─────────────────┐
                    │   Upload Photo  │
                    └────────┬────────┘
                             │
              ┌──────────────┼──────────────┐
              ▼              ▼              ▼
     ┌────────────┐  ┌────────────┐  ┌────────────┐
     │  Reverse   │  │    Face    │  │  Metadata  │
     │   Image    │  │  Analysis  │  │  Forensics │
     │   Search   │  │            │  │            │
     │            │  │  Age       │  │  EXIF      │
     │  Serper    │  │  Gender    │  │  Camera    │
     │  Yandex    │  │  Emotion   │  │  Software  │
     │  G. Lens   │  │  Quality   │  │  GPS       │
     └─────┬──────┘  └─────┬──────┘  └─────┬──────┘
           │   50%         │   30%         │   20%
           └──────────────┬┼───────────────┘
                          ▼
                 ┌─────────────────┐
                 │   Risk Score    │
                 │    0 — 100      │
                 └─────────────────┘

Scoring

Score	Verdict	Meaning
0–30	Likely Safe	Photo appears authentic
30–65	Suspicious	Some red flags detected
65–100	High Risk	Strong indicators of a fake profile photo

What raises the score

Signal	Impact
Found on stock photo sites	+40
Image widely distributed (5+ sites)	+25
Found on social media profiles	+10
EXIF metadata completely stripped	+30
Editing software detected	+20
Overly happy expression (stock pose)	+15
Studio lighting / high symmetry	+10
Single centered face (profile photo)	+10

Setup

Prerequisites

• Python 3.11+
• Node.js 18+
• Serper.dev API key (free — 2,500 searches/month)

Backend

cd backend

# Install dependencies
pip install -r requirements.txt

# Download face analysis models (~120 MB)
python download_models.py

# Configure API key
cp .env.example .env
# Edit .env and add your Serper API key

# Start server
uvicorn main:app --reload --port 8000

Frontend

cd frontend

# Install dependencies
npm install

# Start dev server
npm run dev

Open http://localhost:3000 and upload a photo.

Tech Stack

Frontend

• Next.js 14 (App Router)
• Tailwind CSS + shadcn/ui
• TypeScript

Backend

• FastAPI + Uvicorn
• OpenCV (face detection + DNN age/gender/emotion models)
• Pillow + exifread (metadata extraction)
• Serper.dev API (Google Lens reverse image search)
• Yandex + Google Lens scraping (fallback)

Models

• Age estimation — Caffe DNN (Gil Levi, 2015)
• Gender classification — Caffe DNN (Gil Levi, 2015)
• Emotion recognition — FERPlus ONNX (Microsoft, 2019)
• Face detection — OpenCV Haar Cascade

Project Structure

catfish-detector/
├── backend/
│   ├── main.py                 # FastAPI app + CORS
│   ├── config.py               # Environment settings
│   ├── download_models.py      # Model downloader script
│   ├── models/
│   │   ├── response.py         # Pydantic response schemas
│   │   └── cv_models/          # DNN weights (downloaded via script)
│   ├── services/
│   │   ├── analyzer.py         # Orchestrator (parallel analysis)
│   │   ├── reverse_image.py    # Multi-engine reverse search
│   │   ├── face_analysis.py    # OpenCV + DNN face analysis
│   │   ├── metadata.py         # EXIF extraction
│   │   └── scoring.py          # Weighted score computation
│   ├── routers/
│   │   └── analyze.py          # POST /api/analyze endpoint
│   └── utils/
│       └── image.py            # Validation + resize
│
└── frontend/
    └── src/
        ├── app/page.tsx        # Main dashboard
        ├── components/
        │   ├── image-dropzone.tsx
        │   ├── catfish-gauge.tsx
        │   ├── score-breakdown.tsx
        │   ├── face-analysis-panel.tsx
        │   ├── metadata-panel.tsx
        │   └── reverse-image-panel.tsx
        ├── lib/api.ts          # Backend client
        └── types/analysis.ts   # TypeScript types

License

MIT