Tools are one input, not a verdict
Each tool below sharpens one part of content. Use them to gather evidence, then reason — a single score, label, or marker never settles the question on its own.
AI image detection
These classifiers estimate the probability an image was AI-generated. Treat the score as one weak signal: they are confidently wrong in both directions, especially on compressed, cropped, or re-shared media.
TrueMedia
Nonprofit tool (Georgetown University) that aggregates multiple AI detectors for images and reports uncertainty — built for journalists and the public.
Service availability may vary during relaunch. No detector is perfect; read the uncertainty report, not just the headline score.
Hive Moderation — AI detector
Upload an image or paste a URL to get a likelihood score for AI generation across several model families.
Probabilistic, not definitive. A high or low score should change your confidence a little — never settle the question on its own.
Sightengine — AI image detection
Run a quick check that returns an AI-generated probability; useful as a second opinion alongside reasoning.
Accuracy drops on edited, screenshotted, or low-resolution images.
Illuminarty
Free consumer-facing AI image checker — lower barrier than enterprise demos for a quick first pass.
Consumer-grade accuracy; treat as a prompt to investigate further.
Deepfake / video / audio detection
Video, face-swap, and voice detectors scan for manipulation artifacts. Generation tech keeps outpacing fixed cues, so use these to raise questions, not to close them.
Deepware Scanner
Submit a video or link to scan for face-manipulation/deepfake indicators.
Misses novel methods and can flag authentic video. Pair any result with source and alignment checks.
Sensity AI
Forensic-grade analysis of face manipulation, AI-generated images/video, and synthetic voice — used by investigators and media teams.
Enterprise-focused; some features require an account. Probabilistic output.
Reality Defender
Multi-model deepfake detection for images, video, and audio — upload media or use the API for a quick authenticity check.
Free tier has scan limits. Results are one input among many.
Aurigin.ai
Upload or submit audio to check for AI-generated, cloned, or synthetic voice — useful when a claim hinges on who supposedly said something.
Probabilistic output; novel voice clones may evade detection. Pair with source tracing and independent corroboration.
Metadata & forensics (non-AI)
Before reaching for an AI detector, check whether the file itself tells a story — editing traces, timestamps, and compression patterns often reveal recycled or manipulated media.
FotoForensics
Run Error Level Analysis (ELA) to spot re-compression and editing — regions that were saved at different quality levels stand out.
ELA is interpretive, not definitive; social-media compression creates false positives.
Jeffrey's Exif Viewer
Inspect EXIF metadata: camera model, capture date, GPS, and software used — often stripped on re-upload but worth checking on originals.
Amnesty YouTube DataViewer
Extract upload date, title, and key metadata from YouTube videos to cross-check timing against the claim.
Video forensics & context search
Often the strongest move isn't a detector at all — it's finding where media really came from. Authentic footage is frequently real but recycled from another event, place, or year.
Meet the Amito toolkit
Most verification tools handle only one part of the STOP&SCAN framework — and many aren't free to use. Our program's technical cohort has been building Amito, the AI Media Integrity Toolkit: a conversational bot that turns the framework into an evolving, interactive toolkit you can chat with.
Start a chat with the Amito bot on Telegram.
Open on Telegram
@AMITOAIBOT