Upload an image and get a 0-100 AI-confidence score in seconds - free, and entirely in your browser.
Drag & drop or click to select any photo, artwork, or picture you want to check. Nothing is uploaded - it stays on your device.
The tool checks C2PA provenance and metadata first, then runs an on-device ML model and visual heuristics if needed.
A 0-100 AI-confidence score and verdict appear with the evidence behind it - copy the result to share or save.
Layered detection that reads provenance first and falls back to an on-device model and heuristics - free, private, and honest about its limits.
Every check runs inside your browser. Your image is never uploaded to any server.
Reads Content Credentials (C2PA) manifests when present - the strongest signal a tool like this can use.
Combines provenance metadata, an on-device ML classifier, and pixel-level heuristics into one score.
No sign-up, no credits, no subscription. Check as many images as you like.
Shows a confidence score and evidence, not a false guarantee - AI detection is never 100% certain.
Most images are scored in a couple of seconds once the on-device model is cached.
It layers three signals: C2PA content-credential verification, EXIF/metadata checks for AI software tags, and - if those are inconclusive - an on-device ML model plus pixel-level heuristics, fused into a single 0-100 score.
No detector is 100% accurate. The score reflects the strength of the evidence found, not a certainty - treat it as a strong signal, not proof.
C2PA (Content Credentials) is a tamper-evident manifest some cameras and AI tools attach to images recording how they were made. When present and valid, it gives one of the strongest signals available.
No. Detection runs entirely in your browser using WebAssembly and an on-device model. Your image is never sent to any server.
It looks for provenance and patterns left by common generators - Midjourney, DALL·E, Stable Diffusion, Gemini/Nano Banana and others - through metadata tags and learned visual patterns, though newer or edited images may reduce accuracy.
No. Invisible embedded watermarks aren't read directly; the detector relies on visible metadata, provenance manifests, and visual analysis instead.
