Imposter Shield - AI Image Detector

Flags AI images and videos as you browse, including Shorts, TikTok and Reels. Reads platform AI labels. Runs on your computer.

Imposter Shield tells you when a picture or a video you are looking at was made by AI. It reads the page itself, right on your computer. No account, no uploads, and nothing you browse is ever sent to us. The toolbar icon lights up green, yellow or red depending on what is on the page. WHAT IT CATCHES - Images carrying C2PA Content Credentials, the AI provenance standard - Fingerprints that popular AI image generators leave in the file - Images served straight from AI-only image hosts - Image search results that come from AI image generator sites, or that the stock site they come from labels AI-generated - Search ads and results whose picture leads to an AI image generator - Videos the video platform itself has labelled as AI - Social posts carrying the platform's own "AI content" label - Videos that still show the logo their AI tool stamped on them, even when the platform never labelled them - Full-size pictures with no AI signature at all, judged by a detector that runs on your computer and shown as "Likely AI", never as proof - The full-size originals behind image search results: when a result names where its original file lives, the same detector fetches that picture and takes its own look, shown as yellow, never as proof - Short videos with no label at all: the same detector looks at a few frames of the one you are watching and gives it a red, yellow or green call WHERE PEOPLE USE IT - Marketplace and rental listings with photos of places that do not exist - Dating profiles - Short-video and social feeds - News sites - Product shots and reviews BEING STRAIGHT WITH YOU Imposter Shield keeps proof and judgment apart. A solid outline means the file itself carries an AI signature. Plenty of images have had that stripped by the time they reach you, so for full-size pictures with no signature, a detector on your computer gives its judgment, shown as a dashed "Likely AI" outline. That is never counted as proof, it can turn the icon yellow but never red, and it does not come with a percentage. We tuned it to stay quiet rather than call your own photos fake: on our own test bench it flags about half of current AI images, about a quarter of those from the most popular chatbot, and wrongly flags about 1 in 100 real photos. Pictures that lean AI but fall short get a dotted "Maybe AI" outline, an educated guess that never changes the icon's color. The popup says exactly that. A green light means we found no AI signature, not that we promise the picture is real. We are not going to invent a number to look clever. On video there are three checks. If the platform has labelled it as AI, we say so and tell you who said it. If the video still shows the logo its AI tool stamped on it, we name the tool. And while you watch, the same on-device detector looks at a few frames of the video: sure about more than one of them turns the icon red, some frames leaning AI turns it yellow, clean frames turn it green. That last check is our judgment, not proof, and the popup says so. Some AI video tools stamp their own logo on what they make. When the video you are watching still shows one, Imposter Shield names the tool. It looks for the real logo, not the word, and a moving logo has to move the way the real one does, because people paste fake logos onto real videos as a joke. A watermark can still be faked, so we call it strong evidence, not proof. The frames are checked on your computer and never leave it. Some platforms use one "AI content" label for everything from touching up a photo to generating the whole picture from nothing. So we report that the platform labelled a post, and we tell you exactly that much. We do not turn it into "this is fake", because the platform did not say that. PRIVACY Everything runs locally. There is no account, no advertising, and nothing you browse is ever sent to us. Five things do leave your computer and they are all in our privacy policy in full: the first time it runs, the extension downloads its two detector models (about 110 MB together) from our server once; once a day it reports a random ID created at install plus its version number, so we can count installs and spot a broken release; if you click "This flag is wrong" we receive that one image address and the domain it sat on; when a search result comes from a stock site that labels AI pictures, the extension asks that site for the picture's own page to read its label, with no cookies and nothing about you or your search; and when a search result names its full-size original, the extension may download that picture from the site hosting it so the detector can judge it on your computer, again with no cookies. That is the whole list. Free, and staying free.