AI Detection — checklist operativa
✅ AI Detection Checklist
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.
- Inspect hands: count fingers, check proportions, examine nail shape and placement.
- 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.
- Look for stitching errors, inconsistent textures, or repeating patterns.
- 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.
- Inspect signage and text in the background for legibility.
- 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.
- Ensure all shadows align with a single light source.
- Reflections
- Inspect mirrors, windows, and water surfaces.
- Ensure reflected objects match reality (orientation, size, color).
- Inspect mirrors, windows, and water surfaces.
- 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.
- Run reverse image search on full image and cropped anomalies.
🎥 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.
- Scrub video frame by frame for inconsistencies.
- Motion & Blur
- Validate natural motion blur; AI often generates sharp unnatural edges during motion.
- Look for “halo” effects or ghosting in moving objects.
- Validate natural motion blur; AI often generates sharp unnatural edges during motion.
- Context Checks
- Technical Tools
- Extract thumbnails and keyframes with InVID.
- Run reverse video searches.
- Analyze encoding for unusual compression signatures.
- Extract thumbnails and keyframes with InVID.
- AI Detection
- Run frames through SensityAI or Reality Defender.
- Apply forensic CNNs like FaceForensics++.
- Run frames through SensityAI or Reality Defender.
🔊 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.
- Identify robotic cadence, flat intonation, or overly clean delivery.
- Spectrogram Analysis
- 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).
- Validate background sounds (birds, traffic, wind) against claimed setting.
- 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.
- Run AI voice detection with Deepware Scanner or Intel FakeCatcher.
📝 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.
- Identify repetitive scaffolding (e.g., “In conclusion, …”).
- Factual Validation
- Spot-check quotes and references against primary sources.
- Cross-verify dates, events, and names in OSINT databases.
- Spot-check quotes and references against primary sources.
- Technical Tools
🌍 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).
- Extract with ExifTool.
- Compression & Encoding
- Validate JPEG quantization tables.
- Compare video codecs against known device profiles.
- Validate JPEG quantization tables.
- Sensor Noise & PRNU
- Run Noiseprint for sensor fingerprinting.
- Compare with known authentic samples.
- Run Noiseprint for sensor fingerprinting.
- Provenance Checks
- Check for C2PA metadata.
- Validate Adobe Content Credentials.
- Run Google SynthID watermark checks if available.
- Check for C2PA metadata.
🤝 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.