Skip to content

Quick Start (5 min)

This guide will get you up and running with Veridex in under 5 minutes.

Installation

Veridex is modular. You can install the core library, or include dependencies for specific modalities (text, image, audio, video).

Core Installation

pip install veridex

With Modal Dependencies

pip install "veridex[text]"    # For advanced text signals (transformers)
pip install "veridex[image]"   # For image signals (diffusers, torch)
pip install "veridex[all]"     # Install everything

Text Detection Example

Here's how to check if a text is generated by AI using the PerplexitySignal. This signal analyzes the complexity of the text; AI models often produce "lower perplexity" (more predictable) text than humans.

from veridex.text import PerplexitySignal

# Initialize the signal
# (No arguments needed for default heuristic mode)
signal = PerplexitySignal()

# Analyze text
text = "This is a sample text to analyze. AI often writes very predictable sentences."
result = signal.detect(text)

# Print the result
print(f"AI Probability Score: {result.score:.4f}")
print(f"Confidence: {result.confidence:.4f}")
print(f"Metadata: {result.metadata}")

Image Detection Example

For image detection, you can use signals like DIRESignal (Diffusion Reconstruction Error) or CLIPSignal. Note that these require heavy dependencies (veridex[image]).

from veridex.image import DIRESignal

# Initialize the signal (will verify if torch/diffusers are installed)
try:
    signal = DIRESignal()

    # Path to an image file
    image_path = "path/to/image.jpg"

    result = signal.detect(image_path)

    if result.error:
        print(f"Error: {result.error}")
    else:
        print(f"AI Probability Score: {result.score:.4f}")
        print(f"Confidence: {result.confidence:.4f}")

except ImportError:
    print("Please install image dependencies: pip install veridex[image]")

Understanding the Result

All signals return a DetectionResult object with the following fields:

Field Type Description
score float Probability that the content is AI-generated. Range [0.0, 1.0].
0.0 = Human, 1.0 = AI.
confidence float How reliable the signal considers this specific assessment. Range [0.0, 1.0].
metadata dict Signal-specific metrics (e.g., perplexity values, entropy scores) for debugging or advanced logic.
error str If present, the detection failed. Check this field before trusting the score.

Next Steps