— For enthusiasts & OSINT
Verify before you amplify.
Zoomed-in fingers and “look at the eyes” threads don't scale, and they're wrong more than anyone admits. Score the image with the same calibrated pipeline forensic labs use, in the time it takes to quote-post.
— Use cases
Where a calibrated score beats squinting
Breaking news
Viral image triage
A “photo from the scene” is spreading fast. Get a threshold-calibrated verdict before it enters your thread, your server, or your notes.
Accounts
Avatars & personas
Suspected bot networks love generated portraits. Batch-check profile pictures over the API and rank by primary score.
Communities
Fact-check receipts
Attach the JSON report (hashes, thresholds, limits) instead of a hunch. Your correction carries its own evidence.
Marketplaces
Listings & rentals
Too-perfect product shots and apartment photos that don't exist. One check before you pay a deposit.
Archives
Datasets & collections
Sweep an image folder through the batch endpoint and quarantine what flags at your chosen standard.
Discourse
Settling the argument
“It's obviously AI” vs “it's obviously real”. Replace both with a number, its threshold, and its known error rate.
— Why it works on real posts
Calibrated for the gutter of the internet
By the time an image reaches you it's been re-encoded by WhatsApp, screenshotted on a phone, and re-uploaded twice. Detectors that shine on pristine files fall apart there, so our benchmark degrades every test image through those exact chains and calibrates verdicts per condition.
The journey your evidence took
originalgenerator output or camera filewhatsappmessenger re-encode · 1600px dual passscreenshotphone screenshot · rescale + brightness shiftsocial_chainwhat you finally saved, and what we calibrate for
Detection power under social_chain is printed on every report: 55% at the standard cap on our current run. When the honest number is modest, you deserve to know it.
— Score literacy
Read the score like an analyst, not a fan
The primary score lands between 0 (confident real) and 1 (confident AI). What matters is where it sits relative to the thresholds calibrated for your image's condition, shown here for the social-chain condition.
Below both thresholds. Not cleared, just not flagged. Say “no detection at a 5% false-positive standard”, not “verified real”.
Above the standard threshold for this condition, below strict. A solid triage signal, worth a provenance dig before you amplify.
Above the strict threshold calibrated to ≤1% false positives on every real source. As strong as a single-image signal honestly gets.
— Scriptable
Your tooling, your pipeline
The report is JSON with stable fields. Create a key on your account page, then score images from a script and pipe the result into your sheets, bots, and notebooks. It spends the same quota as the web app.
SHA=$(sha256sum suspicious.jpg | cut -d' ' -f1)
TOKEN=$(curl -s -X POST https://vestigeforensics.com/api/v1/token \
-H "Authorization: Bearer $VF_API_KEY" \
-H "Content-Type: application/json" \
-d "{\"sha256\":\"$SHA\"}" | jq -r .token)
curl -s -X POST https://api.vestigeforensics.com/v1/analyze \
-H "Authorization: Bearer $TOKEN" \
-F "image=@suspicious.jpg" | jq '.decision.verdict'— Further reading
Guides from our benchmark work
Written from published results, limits included.
How to tell if an image is AI-generated
Reverse image search, visual tells, provenance, then calibrated detection: the order that works, what each step misses, and the real numbers from our benchmark.
How accurate are AI image detectors? The measured answer
Vendors say 99%. Our benchmark measures 55 to 64% detection at a 5% false-positive cap, and 36 to 45% at the strict setting. Where the gap comes from.
Is that AI image detector accurate? Judging a vendor's claim
A checklist for any AI detection tool: the four disclosures a real accuracy claim makes, and how to test one in twenty minutes with photos you already have.
AI image detection glossary: forensics terms, plainly defined
Thirty-five terms from AI-image detection and digital forensics in plain language: false positive rate, calibration, base rate, chain of custody.
Check the image before you quote-post it
Free trial with an account: 2 analyses a week, or 100 a month on Vestige Pro. The report is yours to attach, cite, or delete.