A-EYE: AI Image Detector

Was this photo made with AI? A-EYE estimates whether a photo was generated by AI and, when it looks suspicious, marks where the traces are strongest. No account needed: pick a photo or take one. The app is in English and Korean. It follows your phone's language, and you can change it in Settings.

A-EYE won the AX Platform Award (kt cloud CEO Award) at the 2026 K-AI Content Competition, hosted by KT Group as part of Korea's nationwide AI competition.

What it does • AI detection score — upload a photo and get an AI detection score out of 100 (the higher, the closer to AI-generated) with a plain verdict: “Looks like a real photo”, “Hard to tell” or “Looks AI-generated”. • Suspicious areas — for results that suggest AI generation or a partial edit, the area with the strongest traces is highlighted on your photo. • Partial edits — when a photo looks real overall but part of it shows strong traces of editing, A-EYE says so separately (“May be partly edited”). • It says when it is not sure — borderline photos are reported as “Inconclusive” instead of being forced into a yes or no. • AI records in the file — if the file's C2PA provenance record says it was made, or edited, with AI, A-EYE reports that. Other AI marks in the file info or file name can be written in by anyone, so on their own they never make a result AI-generated; at most, judgment is withheld (“Hard to tell”). Capture records written by a camera are not treated as AI. • History and sharing — your history stays on your device, and you can share a result as an image.

Please keep in mind • Results are statistical estimates, not proof. A real photo can be judged as AI, and an AI image as real. • Images from very recent generators, images compressed by repeated forwarding, and photos of a screen are harder to judge. • Do not use a result as the only basis for legal evidence, news reporting, hiring or similar decisions.

Privacy • No account, no ads, no tracking. • Only the photo you choose is sent to the analysis server over an encrypted connection, and it is not kept after the analysis. It is not used for model training. • The camera permission is used only when you take a photo. The app never accesses your whole photo library.

The analysis runs on a server, so an internet connection is required. A result usually takes about 10 seconds; the first analysis after a while can take 1–3 minutes while the server gets ready.

Discovery signals