Forensic Image Analysis
Detect Deepfakes
with AI Precision
10 forensic algorithms + Claude AI vision. Results in seconds — all pixel processing on your device.
Two-Layer Detection
Client algorithms + AI vision
10 forensic methods run entirely in your browser — no image data leaves your device. If the score is ambiguous, Claude AI Vision examines subtle visual artifacts that pixel math cannot detect.
Forensic Methods
Error Level Analysis
Compares JPEG re-compressions; AI images show abnormally flat or uniform ELA patterns.
Noise Analysis
Laplacian high-pass filter + patch uniformity. GAN outputs have near-zero noise floors.
Frequency Analysis
Autocorrelation at lags 8/16/32. Convolutional upsampling leaves periodic spatial artifacts.
Color Coherence
Hue entropy + saturation variance. Diffusion models produce suspiciously narrow palettes.
Edge Integrity
Sobel edge density CoV. Face-swap masks flatten local sharpness variation unnaturally.
Metadata Presence
Scans binary headers for EXIF/XMP. Real photos carry device data; AI images rarely do.
Texture Complexity
Local Binary Pattern entropy. AI images often show homogeneous, low-entropy texture fields.
Sharpness Uniformity
Patch-level variance map. Diffusion models lack real lens blur and focus falloff.
Channel Statistics
RGB skewness and kurtosis. Synthetic images deviate from natural photo distributions.
Symmetry Analysis
Bilateral pixel symmetry deviation. GAN faces show atypical left-right symmetry profiles.
Limitation notice: Heuristic forensic techniques — no method achieves perfect accuracy. Results are indicative, not conclusive. Do not use to make real-world judgements about individuals.