Skip to content

Veridex Documentation

Veridex Logo

A modular, probabilistic, and research-grounded AI content detection library

Welcome

Veridex is a production-ready library for detecting AI-generated content across multiple modalities: text, image, audio, and video. Unlike binary classifiers, Veridex provides probabilistic detection with confidence estimates and interpretable signals.

🎯 Multi-Modal

Detect AI-generated text, images, audio, and video with a unified API

📊 Probabilistic

Get confidence scores and probabilities, not just binary labels

🔍 Interpretable

Access individual detection signals for transparency

🧩 Modular

Easy to extend with new detection methods

🚀 Production-Ready

Robust error handling and graceful degradation

📖 Research-Grounded

Based on state-of-the-art academic papers

Quick Start

Get started with Veridex in under 5 minutes:

from veridex.text import PerplexitySignal

# Initialize detector
detector = PerplexitySignal()

# Analyze text
result = detector.run("Your text here...")

print(f"AI Probability: {result.score:.2f}")
print(f"Confidence: {result.confidence:.2f}")

from veridex.image import FrequencySignal

# Initialize detector
detector = FrequencySignal()

# Analyze image
result = detector.run("image.png")

print(f"AI Probability: {result.score:.2f}")

from veridex.audio import SpectralSignal

# Initialize detector
detector = SpectralSignal()

# Analyze audio
result = detector.run("audio.wav")

print(f"AI Probability: {result.score:.2f}")

from veridex.video import VideoEnsemble

# Initialize detector
detector = VideoEnsemble()

# Analyze video
result = detector.run("video.mp4")

print(f"AI Probability: {result.score:.2f}")
print(f"Confidence: {result.confidence:.2f}")

📚 Full Installation Guide →

How It Works

graph LR
    A[Input Data] --> B{Modality}
    B -->|Text| C[Text Signals]
    B -->|Image| D[Image Signals]
    B -->|Audio| E[Audio Signals]
    B -->|Video| F[Video Signals]
    C --> G[Normalization]
    D --> G
    E --> G
    F --> G
    G --> H[Detection Result]
    H --> I[Score + Confidence]

Veridex uses a signal-based architecture where each detector:

  1. Analyzes specific features or artifacts
  2. Returns a standardized DetectionResult
  3. Operates independently for transparency
  4. Declares its limitations explicitly

🏗️ Learn More About Architecture →

Choose Your Path

🚀 Quick Start
Get running in 5 minutes 📝 Text Detection
Analyze LLM outputs 🖼️ Image Detection
Spot AI-generated images 🎵 Audio Detection
Detect voice deepfakes 🎥 Video Detection
Detect deepfake videos

Available Detectors

Text Detection

Signal Method Speed Accuracy Best For
PerplexitySignal Statistical (LLM) Medium ⭐⭐ General purpose
BinocularsSignal Contrastive Medium ⭐⭐⭐ High accuracy
ZlibEntropySignal Compression Fast ⭐ Quick screening
StylometricSignal Linguistic Fast ⭐ Style analysis

Image Detection

Signal Method Speed Accuracy Best For
FrequencySignal Spectral Fast ⭐⭐ Quick screening
DIRESignal Diffusion Slow ⭐⭐⭐ High accuracy
ELASignal Error Level Fast ⭐⭐ Manipulation detection

Audio Detection

Signal Method Speed Accuracy Best For
SpectralSignal Frequency Fast ⭐⭐ Lightweight
AASISTSignal Spectro-Temporal Medium ⭐⭐⭐ Anti-spoofing
Wav2VecSignal Foundation Model Slow ⭐⭐⭐⭐ Production
SilenceSignal Pause Analysis Fast ⭐ Speech patterns

Video Detection

Signal Method Speed Accuracy Best For
RPPGSignal Biological Medium ⭐⭐⭐ Face swaps
I3DSignal Spatiotemporal Medium ⭐⭐⭐ General deepfakes
LipSyncSignal Audio-Visual Medium ⭐⭐⭐ Dubbing
VideoEnsemble Fusion Medium ⭐⭐⭐⭐ Production use

📦 See All Detectors →

📖 Tutorials

Step-by-step guides for each modality

Browse Tutorials →

🔍 Concepts

Understand the architecture and signals

Learn Concepts →

📘 API Reference

Complete API documentation

View API Docs →

❓ FAQ & Troubleshooting

Common questions and solutions

Get Help →

🎯 Use Cases

Real-world applications

Explore Use Cases →

⚡ Performance

Optimization and benchmarking

Optimize Performance →

What's New

Latest Updates

  • ✅ Added SilenceSignal for audio detection
  • ✅ Added StylometricSignal for text analysis
  • ✅ Added ELASignal for image manipulation detection
  • ✅ Added Video Detection support (RPPG, I3D, LipSync)
  • ✅ Improved test coverage to 80%+
  • ✅ Enhanced documentation with tutorials

📝 View Full Changelog

Community & Support

Have questions or want to contribute?

Connect with Me: Aditya Mahakali on LinkedIn

Built with ❤️ for transparency in the age of generative AI