stegmark 1.0.3

Universal steganographic analysis — statistical, forensic, and neural watermark detection

Stegmark

Universal steganographic analysis — statistical, forensic, and neural watermark detection.

Install

pip install stegmark                  # statistical + forensic detectors
pip install "stegmark[neural]"        # + TrustMark / Stable Signature neural detection

Usage

from stegmark import analyze, robustness_test

# Analyze any image
result = analyze("image.jpg")
print(result.verdict)
print(result.confidence)
result.save_report("audit.pdf")

# Test watermark robustness
report = robustness_test("cover.jpg", secret="my_id")
print(f"Survived {report.survival_rate:.0%} of attacks")
print(f"Failed: {report.failed_attacks}")
report.save_report("robustness_audit.pdf")

# With cover image for quality metrics
result = analyze("stego.jpg", cover_path="original.jpg")
print(f"PSNR: {result.psnr:.1f} dB  SSIM: {result.ssim:.4f}")

What it detects

Detector	Domain	Catches
Chi-square	JPEG DCT	LSB replacement in DCT coefficients
RS Analysis	Spatial	Pixel-level LSB substitution
SPA	Spatial	Sample pair analysis
EOF Detection	Forensic	Data appended after image end marker
EXIF Analysis	Forensic	Metadata anomalies, steg tool fingerprints
LSB Histogram	Forensic	LSB plane uniformity
Multi-channel	Forensic	Per-R/G/B channel embedding
TrustMark	Neural	Adobe C2PA robust watermarks
Stable Signature	Neural	Meta AI Stable Diffusion watermarks
SRNet	Neural/GPU	Adaptive spatial steg (WOW/S-UNIWARD/HILL)

API

Also available as a hosted API: https://raghav7006--stegmark-api-web.modal.run

curl -X POST https://raghav7006--stegmark-api-web.modal.run/analyze \
  -F "stego=@image.jpg"

Installation Notes

Apple Silicon (M1/M2/M3 Mac)

jpegio requires a native arm64 build:

ARCHFLAGS="-arch arm64" pip install jpegio
pip install stegmark

Linux

pip install stegmark

Windows

jpegio is not supported on Windows. Use WSL2 or Docker.