Public checks
Every check
we publish.
Images and videos from public URLs, run through the same detector as the free checker, with the result and its limits kept next to it. 3 items checked so far.
These pages were checked with older released detectors, before the current V10 routed lane (aidetect-v16-image-384-v1+owenslab+siglip-probe-routed-v11+dualview-orig-v1). They show how the published model behaved at the time of the check; the accuracy page carries the current measured state.
What is in here.
Only media reachable from a public link — a video, a post, a file anyone can open. Checks published here are ours, run deliberately on public sources. Nothing a visitor uploads to the free checker appears here, ever: those files stay temporary, as the privacy notice says.
How to read these pages.
Each page reports what the detector measured, not what happened. A strong signal means the image looks like the output of a generator this detector knows; it does not establish who made something, which tool was used, or that anyone set out to mislead.
Measured through the deployed pipeline: 78.1% strong warnings on 653 AI images, 4.0% false warnings on 795 real ones. These are bounded slices, not an overall accuracy figure and not proof of origin. The false alarms sit mostly on human-made graphics and logos. All of it is measured and published, including the failures.
If you believe a result on this page is wrong, tell us and we will recheck it and correct the page.