AI Detection — checklist operativa

✅ AI Detection Checklist

Version
Last Update

A comprehensive and detailed checklist for detecting and analyzing AI-generated or manipulated media — including images, video, audio, and text. It is intended for OSINT investigators, journalists, digital forensic analysts, and security professionals who require structured, reliable verification procedures.

🖼️ Image & Visual Content Verification

  • Anatomy & Object Integrity
    • Inspect hands: count fingers, check proportions, examine nail shape and placement.
    • Examine eyes: look for mismatched irises, unnatural reflections, symmetry that is “too perfect.”
    • Inspect teeth: check if rendered as a uniform block, misaligned gum lines, or blurry interiors.
    • Inspect ears and earrings: check for asymmetry, distortions, or blending into hair/skin.
    • Check accessories (glasses, hats, jewelry) for warped edges, melting, or blending artifacts.
  • Clothing & Fabrics
    • Look for stitching errors, inconsistent textures, or repeating patterns.
    • Validate logos and printed text on clothing (AI often renders gibberish or blurred letters).
    • Check folds and shadows in fabric for natural consistency.
  • Background Consistency
    • Inspect signage and text in the background for legibility.
    • Identify warped objects (lamp posts, buildings, cars).
    • Check perspective lines: ensure vanishing points are consistent.
  • Lighting & Shadows
    • Ensure all shadows align with a single light source.
    • Validate intensity and direction of light on different objects.
    • Use SunCalc to validate time of day.
  • Reflections
    • Inspect mirrors, windows, and water surfaces.
    • Ensure reflected objects match reality (orientation, size, color).
  • Technical Checks
    • Run reverse image search on full image and cropped anomalies.
    • Perform Error Level Analysis (ELA) via Forensically.
    • Inspect for cloning or copy-paste elements.
    • Review EXIF metadata using ExifTool.
    • Validate GPS and timestamps against claimed context.

🎥 Video Verification

  • Frame-by-Frame Analysis
    • Scrub video frame by frame for inconsistencies.
    • Check lips vs. phonemes for sync accuracy.
    • Look for flickering artifacts or blending errors.
  • Motion & Blur
    • Validate natural motion blur; AI often generates sharp unnatural edges during motion.
    • Look for “halo” effects or ghosting in moving objects.
  • Context Checks
    • Verify landmarks, signs, and clothing seasonality.
    • Cross-check weather with Meteostat or OGIMET.
    • Use SunCalc for shadow analysis.
  • Technical Tools
    • Extract thumbnails and keyframes with InVID.
    • Run reverse video searches.
    • Analyze encoding for unusual compression signatures.
  • AI Detection
    • Run frames through SensityAI or Reality Defender.
    • Apply forensic CNNs like FaceForensics++.

🔊 Audio Verification

  • Listening Checks
    • Identify robotic cadence, flat intonation, or overly clean delivery.
    • Check for missing human sounds (breathing, mouth clicks, filler words).
    • Detect looping or repetitive background noise.
  • Spectrogram Analysis
    • Generate spectrograms in Audacity or Praat.
    • Look for:
      • Clean unnatural high frequencies.
      • Banding artifacts.
      • Missing natural harmonics.
  • Environmental Verification
    • Validate background sounds (birds, traffic, wind) against claimed setting.
    • Compare ambient audio with expected acoustics (e.g., indoor echo vs. outdoor open field).
  • Technical Tools
    • Run AI voice detection with Deepware Scanner or Intel FakeCatcher.
    • Compare with known samples of the speaker.
    • Analyze jitter/shimmer metrics in Praat.

📝 Textual Verification

  • Linguistic Checks
    • Identify repetitive scaffolding (e.g., “In conclusion, …”).
    • Look for vague or generic phrasing without specifics.
    • Check for fabricated citations, URLs, or ISBNs.
  • Factual Validation
    • Spot-check quotes and references against primary sources.
    • Cross-verify dates, events, and names in OSINT databases.
  • Technical Tools
    • Run GLTR or DetectGPT.
    • Perform stylometric comparison with JStylo.
    • Use HuggingFace AI detection models for second opinion.

🌍 Contextual & Environmental Consistency

  • Validate location using Google Earth and Street View.
  • Cross-check building architecture with regional styles.
  • Verify vegetation/season (trees in bloom vs. claimed season).
  • Validate weather data with Meteostat or OGIMET.
  • Check holidays, political events, or known gatherings on claimed date.
  • Match crowd size against known venue capacity.

📊 Metadata & Technical Fingerprints

  • EXIF Analysis
    • Extract with ExifTool.
    • Look for missing or improbable fields.
    • Detect editing software tags (Stable Diffusion, MidJourney, Photoshop).
  • Compression & Encoding
    • Validate JPEG quantization tables.
    • Compare video codecs against known device profiles.
  • Sensor Noise & PRNU
    • Run Noiseprint for sensor fingerprinting.
    • Compare with known authentic samples.
  • Provenance Checks
    • Check for C2PA metadata.
    • Validate Adobe Content Credentials.
    • Run Google SynthID watermark checks if available.

🤝 Peer Review & Validation

  • Share findings with a second analyst for independent validation.
  • Compare across multiple tools and detection methods.
  • Store SHA256 and MD5 hashes of original files for integrity.
  • Maintain chain of custody logs for evidentiary purposes.
  • Document all anomalies with annotated screenshots.
  • Record all commands, queries, and tools used for auditability.

🛠️ Quick Access Tools

Tool Type Purpose
Forensically Image forensics Error level analysis, clone detection, metadata review
InVID Plugin Video verification Extract thumbnails, reverse video search, metadata analysis
ExifTool Metadata extraction Inspect and validate EXIF and file metadata
GLTR Text analysis Detect statistical patterns in AI-generated text
DetectGPT Text analysis Identify likely AI-generated passages
Noiseprint Image forensics Sensor fingerprinting and camera source validation
Audacity Audio analysis Waveform and spectrogram inspection
Praat Audio forensics Acoustic analysis of speech and voice patterns
Meteostat Contextual data Historical weather validation
SunCalc Contextual analysis Validate shadows and sun positions by time and place
SensityAI AI detection Deepfake and synthetic media detection services

🔖 Credits

Maintained by Oryon + OSINT360.
This document is part of the Cyber Intelligence Toolkit project.

Themes