Processing...

AI Image Screener

A practical first-pass AI image screening system designed to identify images that require human review based on statistical and physical patterns.

Screening accuracy: 40-90% detection rate across AI models

Fast Processing

Parallel processing for batch analysis with real-time progress tracking and optimized performance.

Multi-Signal Detection

Five independent statistical detectors with weighted ensemble aggregation for reliable results.

Evidence Analysis

Aggregates detection signals and metadata into structured evidence, resolving conflicts.

Decision Policy

Applies deterministic rules over metrics and evidence for review-aware final verdict.

Comprehensive Reports

Export results in CSV and JSON formats for integration, documentation, and audit trails.

Important Notice

This is not a perfect AI detector. It's a screening tool that helps reduce manual review workload by flagging suspicious images for human verification. Always use human judgment for final decisions.

Signal-based Detection Metrics
Statistical analysis of image properties with weighted scoring
Gradient-Field PCA
30%

Detects lighting & gradient inconsistencies typical of diffusion models. Analyzes directional light patterns and shadow consistency.

Method Principal Component Analysis
Detection Rate 85-95%
Sensitivity High
Frequency Analysis
25%

Identifies unnatural spectral energy distributions via FFT analysis. AI-generated images often show characteristic frequency patterns.

Method Fast Fourier Transform
Detection Rate 75-85%
Sensitivity Medium-High
Noise Pattern
20%

Detects missing or artificial sensor noise patterns. Real cameras produce characteristic noise while AI models generate uniform patterns.

Method Noise Distribution
Detection Rate 70-80%
Sensitivity Medium
Texture Statistics
15%

Identifies overly smooth or uniform texture regions. AI-generated images often lack natural texture variation found in real photographs.

Method GLCM Texture
Detection Rate 60-70%
Sensitivity Medium-Low
Color Distribution
10%

Flags unnatural saturation and color histogram patterns. AI models produce colors with distribution patterns different from real photographs.

Method Color Histogram
Detection Rate 50-65%
Sensitivity Low-Medium
Evidence-based Verification
Additional verification through metadata and watermark analysis
EXIF Analyzer
Evidence

Analyzes image metadata for presence, completeness, and plausibility. Real camera images contain coherent EXIF data while AI-generated images often lack or have inconsistent metadata.

Method Metadata Analysis
Key Signals Missing/Inconsistent EXIF
Verification Medium
Watermark Analyzer
Evidence

Detects known and statistical watermark patterns embedded by generative models. Includes checks for frequency-domain artifacts and spatial regularities associated with AI watermarking techniques.

Method Pattern Analysis
Key Signals Watermark Artifacts
Verification Model-Dependent
1

Upload Images

Drag & drop or select multiple images (JPG, PNG, WEBP formats supported).

2

Start Analysis

Click "Start Analysis" to begin screening with real-time progress tracking.

3

Review Results

Check flagged images, view detailed analysis, and export comprehensive reports.