ReMorph – Tracing the origin of deception

ReMorph is a forensic AI tool that goes beyond detection. It doesn’t just flag deepfake images — it traces them back to their likely source.

Deepfake detection alone isn’t enough. Once an image is flagged as synthetic, investigators need to ask:

📌 Who made it? 📌 Which tool was used? 📌 How was it generated?

ReMorph is the first step toward true deepfake traceability, enabling investigators, journalists, and platforms to:

• 💥 Attribute fake content to specific generative models
• 💥 Uncover potential creators or toolchains
• 💥 Strengthen legal and platform-based takedown processes

---

🚀 Features

• 🧠 Deepfake Detection: Flags synthetic media with high confidence
• 🔍 Model Attribution: Identifies the likely generative model (StyleGAN, DeepFaceLab, etc.)
• 📊 Dataset Estimation: Infers dataset styles (celebs, scraped social media, etc.)
• ⚡ Artifact Fingerprinting: Spots hidden traces (checkerboard noise, boundary mismatches, color anomalies)
• 🖼️ Grad-CAM Explanations: Visual heatmaps highlight manipulated regions
• 📑 Forensic Reports: Generates clear reports with origin likelihood scores and metadata
• 🌐 Web Interface: Upload media through a friendly frontend
• 📡 API Endpoints: Swagger UI for programmatic interaction
• 🔒 Privacy-First: Runs locally — no data leaves your machine

Ideal Use Cases

• Investigative Journalism: Verifying suspect media
• Social Media Platforms: Flagging and tracing manipulated uploads
• Legal & Law Enforcement: Building admissible forensic evidence
• AI Governance: Auditing generative models and enforcing accountability

---

ReMorph is built to excel across several evaluation categories:

Criteria	Description
💡 Real-World Impact	Solves the urgent problem of untraceable deepfakes
🛠️ Technical Execution	Built on robust ML + forensic techniques
🎨 Creativity & Innovation	First tool to focus on traceability, not just detection
🎥 Presentation	Clear, accessible UI + reports for non-technical users

---

🧪 Tech Stack

• Backend: Python, FastAPI, Uvicorn, Typescript
• AI/ML: TensorFlow, EfficientNet, CNN classifiers, FFT/DCT analysis
• Frontend: React + Vite, Next.js, Tailwind CSS
• Visualization: Grad-CAM, frequency plots
• Environment: Node.js, Python venv, local server

---

📖 How It Works

1. Upload Media (image now, video support in roadmap)

2. AI Pipeline Runs

   • Detects synthetic artifacts
   • Classifies likely generative model
   • Estimates dataset fingerprints
   • Highlights manipulation regions with Grad-CAM

3. Output

   • Forensic likelihood scores
   • Visual artifact maps
   • Downloadable PDF/HTML reports

---

🏗️ Architecture Overview

ReMorph follows a modular architecture designed for scalability and clarity:

                ┌──────────────────────┐
                │      Frontend        │
                │  React + Tailwind    │
                │  User upload & UI    │
                └─────────┬────────────┘
                          │
                          ▼
                ┌──────────────────────┐
                │      Backend         │
                │   FastAPI + Uvicorn  │
                │  Routes & API logic  │
                └─────────┬────────────┘
                          │
                          ▼
                ┌──────────────────────┐
                │    ML Pipeline       │
                │ TensorFlow + FFT/DCT │
                │ Detect & Attribution │
                └─────────┬────────────┘
                          │
                          ▼
                ┌──────────────────────┐
                │   Report Generator   │
                │   PDF + HTML output  │
                │   Grad-CAM visuals   │
                └──────────────────────┘

• Frontend → Provides a user-friendly interface for uploading and visualizing results.
• Backend → Handles requests, routes, and communication between UI & ML pipeline.
• ML Pipeline → Performs detection, attribution, artifact analysis.
• Report Generator → Produces forensic reports for judges, journalists, and investigators.

---

⚡ Setup Instructions

Clone the repo:

git clone https://github.com/Kehn-Marv/Remorph.git
cd Remorph

Prepare directories (important!)

Before starting, make sure the following folders exist in backend folder root:

mkdir -p outputs
mkdir -p weights

• outputs/ → Stores generated analysis reports and visualizations
• weights/ → Stores ML model weights (downloaded or pre-trained)

Backend

python -m venv .venv
call .venv\Scripts\activate    # Windows
cd backend
pip install -r requirements.txt
python -m uvicorn src.api.main:app --host 0.0.0.0 --port 8080 --reload

👉 Access API at:

• Swagger UI: http://127.0.0.1:8080/docs

Frontend

cd frontend
npm install
npm run dev

👉 Access web app at:

• http://localhost:3000

---

📅 Roadmap

• ✅ MVP: Image detection + attribution
• 🔜 Support for video deepfakes (.mp4, .webm)
• 🔜 Bias & manipulation warnings (age/gender/race edits)
• 🔜 Stronger open-set anomaly detection
• 🔜 Collaborative fingerprint database for new models
• 🔜 Deployable cloud + on-premise versions

---

👥 Authors

• Kehn Marv – GitHub
• Essie – GitHub

---

🆘 Troubleshooting

Backend not starting:

• Ensure Python ≥ 3.10
• Activate venv before installing requirements

Frontend build errors:

• Delete node_modules and reinstall (npm install)
• Ensure Node.js ≥ 18

Swagger UI not loading:

• Check that Uvicorn server is running on port 8080

---

Built with ❤️ for truth, accountability, and digital integrity.