FAQ

Run it through the detector and read the evidence. Signed provenance (C2PA, watermarks) is near-conclusive; metadata hits (prompt chunks, generator fields) are strong; everything else is probabilistic. Also train your own eye on the voting square — real photos often have tell-tale noise and lighting consistency that AI images smooth away.

Yes — it runs entirely in your browser-facing web app, needs no install and no account, and gives you free checks every day. Signing in only unlocks extras: voting, publishing, daily credits, and saved history.

No AI-image detector is 100% accurate — including human eyes (~62% in tests). We rank decisive evidence (digital signatures, watermarks) first: a hit gives a high-confidence verdict. When nothing is found we honestly say "inconclusive" instead of forcing a binary answer.

Social platforms compress images and strip metadata and watermarks. Screenshots and re-saved files lose most detection signals, so the verdict more often lands on "inconclusive".

Detection needs no login (a few free checks per day). Signing in unlocks voting and publishing, daily credits (the classifier second opinion), saved history, and your stats.

Before voting you see neither the counts nor the machine verdict — that keeps community judgment independent. After voting everything is revealed: distribution, machine comparison, and your streak.

Private detection images are auto-deleted after 24 hours. Only images you actively publish to the voting square stay public, and you can withdraw them anytime. See the privacy policy.

A real photo touched up with generative fill (e.g. Photoshop) is "AI-edited", not "AI-generated". C2PA credentials distinguish the two, and our report labels them separately.

Six evidence-graded steps: C2PA signature verification, invisible watermark decoding, EXIF scan, PNG text chunks, IPTC declarations, and forensic signals. See the full methodology for details and known limitations.