GenDetective: AI Content Detector

GenDetective is a multi-modal AI content detector that uses statistical and stylometric analysis as well as LLM reasoning to identify AI-generated images, videos, and text.

Comment

Inspiration

The rapid growth of generative AI tools has made it increasingly difficult to distinguish between human-created and AI-generated digital content. Images, videos, and text produced by modern AI models are now widely shared online without disclosure, raising serious concerns about:

  • Misinformation
  • Academic integrity
  • Fake news
  • Digital trust

This project was inspired by a simple real-world question:

“Is the content I am seeing real or AI-generated?”

To address this problem, we built GenDetective, a multimodal AI-content detection system available as both:

  • A Chrome browser extension
  • A Web dashboard

What it does

GenDetective analyzes images, videos, and text and estimates whether the content is AI-generated or human-created.

The system provides:

  • Classification

    • AI-generated
    • Likely AI
    • Likely real
    • Real
    • Inconclusive

  • Confidence Score

  • Human-readable Explanation

  • Forensic Indicators used in the decision-making process

Users can:

  • Upload media through the browser extension
  • Use the dashboard interface
  • Submit text for analysis

The goal is to make AI detection transparent, explainable, and accessible.

How we built it

GenDetective consists of three main components:

1. Browser Extension

Technologies used:

  • HTML
  • CSS
  • JavaScript

The extension allows users to quickly submit online content for analysis directly from the browser.

2. Dashboard Interface

The web dashboard provides a complete interface for interacting with the detection system.

Features include:

  • Image upload and detection
  • Video upload and detection
  • Text analysis
  • Visualization of detection results

3. FastAPI Backend

The backend processes media and performs the AI detection analysis.

Technologies used:

  • FastAPI
  • NumPy
  • OpenCV
  • Transformers (CLIP)
  • Gemini API
  • scikit-learn

The system uses an ensemble detection approach that combines:

  • Heuristic analysis
  • Machine learning models
  • Embedding similarity
  • LLM-based reasoning

Challenges we ran into

During development, we encountered several technical challenges:

  • Avoiding false positives in compressed media
  • Detecting high-quality AI-generated videos
  • Handling large video uploads efficiently
  • Combining heuristic detection with LLM reasoning
  • Maintaining performance inside a browser extension
  • Working within limited API usage constraints

Accomplishments that we're proud of

  • Built a working multimodal AI detection system
  • Created a Chrome extension connected to a backend
  • Built a dashboard interface for detection
  • Implemented image, video, and text detection pipelines
  • Designed an explainable AI detection workflow
  • Successfully integrated CLIP + Gemini + ML models

What we learned

Throughout the project, we gained several important insights:

  • AI-generated content differs from human content in subtle statistical patterns, rather than obvious visual artifacts.

Key learnings include:

  • No single feature can reliably detect AI content
  • Detection works best by combining multiple weak signals
  • Video detection requires temporal analysis
  • Browser extensions require careful performance optimization

What's next for GenDetective

Future improvements planned for the system include:

  • Expanding benchmarking datasets
  • Improving deepfake video detection accuracy
  • Adding frame-level voting detection
  • Improving motion realism modeling
  • Strengthening text detection robustness
  • Deploying the backend to cloud infrastructure
  • Publishing the browser extension
  • Optimizing dashboard UX and performance

GenDetective aims to help restore trust in digital content by making AI-generated media detectable, explainable, and transparent.

Built With

Share this project:

Updates