> Five checks for a face you suspect is AI-made: what to look at, what a watermark check proves, and how comparing their photos exposes a fake. To tell if a photo of a person is AI generated, do not rely on one clue: combine a close look at the image, a watermark check, an image search, a face search and a comparison of the person's own photos with each other. Each check is weak alone, but together they usually give you a clear enough answer to decide what to do next. This matters most when the face belongs to someone talking to you: a match, a new follower, a recruiter or a seller. **A generated face is built to have no past, so the most useful question is not "does it look fake" but "does this face exist anywhere else, and is it the same face in every photo".** ## LOOK AT THE FACE AND WHAT SURROUNDS IT Start with the image itself. Open it at full size, not as a thumbnail, and zoom in. - **Ears and earrings.** Generators often draw two ears that do not match, or an earring on one side that has no partner. - **Glasses.** Look where the arms meet the head. They may bend into the skin or vanish into hair. - **Hair edges.** Strands that melt into the background or into a collar are a common weak spot. - **Teeth and eyes.** Teeth that merge into one band, or reflections in the eyes that point in different directions. - **The background.** Vague blurred shapes, text that reads as nonsense, objects that do not quite resolve. - **Hands, if visible.** Fused or oddly bent fingers still turn up. Then be honest about the limit. Newer generators make far fewer of these mistakes, and a real photo can have odd-looking details too, especially after a beauty filter or heavy compression. A clean-looking face is not proof of anything. Treat this step as a reason to look harder, not as a verdict. ## CHECK FOR A WATERMARK, AND KNOW WHAT IT CANNOT PROVE Some AI tools mark what they produce. Google's own images and videos carry an invisible SynthID watermark, and Google lets you check for it in the Gemini app: upload one image and ask "Is this AI-generated?". The steps and limits are on [Google's SynthID verification help page](https://support.google.com/gemini/answer/16722517). Read the result carefully: - **Watermark found** means all or part of the image was created or edited by Google's AI models. - **No watermark found** only means it was not made or edited by Google AI. It can still come from any other generator. Google states that the check currently recognises only content made with Google AI tools. It also advises cropping screenshots tightly and not uploading collages. **A missing watermark is not evidence that a face is real.** Most screenshots of a profile picture will tell you nothing here, so move on to the checks that look at where the face has been. ## SEARCH THE IMAGE Run the photo through a free image search such as Google Lens. This looks for the file and near-copies of it. If the same picture turns up on a gallery of generated portraits, a stock library or another profile with a different name, you have your answer: the person messaging you did not take it. If nothing comes back, that is common. A freshly generated face has never been published, so an image search has nothing to find. The [difference between image search and face search](https://traceaifacescan.app/blog/reverse-image-search-vs-face-search/) is why the next step exists. ## SEARCH THE FACE A face search asks a different question: does this same face appear anywhere on the public web, in any photo. Crop the face from the screenshot, keep it front-on and sharp, and [start a search](https://traceaifacescan.app/panel). Trace matches the face rather than the file, so it can find the same person in a completely different photograph. Each result is a public page with a similarity score from 50 to 100 and a link you open yourself. What the outcomes tell you: - **The same face under another name or on several unrelated profiles.** The picture is being reused, whether it was generated or stolen. Either way, it is not this person's own photo. - **The same face on a consistent history.** Older posts, a team page, tagged photos with friends, all under the name you were given. A generated face does not build that kind of trail, so this is reassuring, though it is still a lead rather than proof. - **Nothing at all.** An empty result means nothing was found on the part of the public web Trace can reach. It does not prove the face is generated, and it does not prove the person is genuine. Plenty of real people keep a low profile. Weigh it against the story you were told. ## COMPARE THEIR PHOTOS WITH EACH OTHER This is the check most guides skip. A real person's photos are all the same face, even years apart. A set of generated images may not be: the face can drift between pictures in the jaw, nose or eye spacing, because each picture was produced separately. Take two of the photos the person has shared and put them into [Compare two faces](https://traceaifacescan.app/compare-faces/). It is free, searches nothing and stores nothing, and returns one similarity score on the same scale. Roughly, 70 to 80 is a possible match worth a second look, 80 to 90 is strong and 90 or above is near-certain. If two pictures that are supposed to show the same person score low, ask why. Bad lighting, a side angle or a filter can pull a score down, so try another pair before you conclude anything. ## ASK FOR SOMETHING A GENERATOR CANNOT PREPARE The checks above look backwards. A live request tests the present. Ask for a photo of something specific that nobody could have prepared: a hand making a particular gesture next to the face, or today's date written on paper. Better still, a short video call. A real person can do it in a minute. Repeated excuses tell you more than any detector. ## IF YOU CONCLUDE THE FACE IS GENERATED - **Do not send money, codes or documents.** A generated face on a stranger's profile is a strong sign of a scam, whatever the story. - **Keep the evidence.** Screenshot the profile, the photos and the conversation before you say anything. - **Report the account to the platform** through its own reporting tool, then block it. - **Run the wider checks** in [Is This Person a Scammer?](https://traceaifacescan.app/blog/is-this-person-a-scammer/) if money has already come up, and see the [online dating safety check](https://traceaifacescan.app/blog/online-dating-safety-check/) before you meet anyone from an app. If money has already been lost, contact your bank first and then your national fraud-reporting service or the police. ## RESPONSIBLE USE Use these checks to protect yourself: to confirm that someone talking to you is the person in their photos. Trace is not an identification service. A similarity score compares two images and never states who somebody is; relatives and lookalikes can score high. Trace returns no names, addresses or phone numbers, only public pages and a score. If you would rather your own face were not in Trace's index, you can have it deleted and blocked at [Remove my photos](https://traceaifacescan.app/remove-my-photos), with no account and no ID. ## FAQ ### Can you tell if a picture is AI generated for free? Yes, mostly. Looking closely at the image, checking for a SynthID watermark in Gemini, running a free image search and comparing two of the person's photos with Trace's free face comparison all cost nothing. None of them is conclusive alone, so use several together. ### Is there a website to tell if a photo is AI generated? Google's Gemini app can check for its own SynthID watermark, but only content made with Google AI tools will show one. Other detector sites give a probability, not a fact. Treat any single tool's answer as one input next to the image search, the face search and a live request. ### Can a reverse image search detect AI images? Not directly. It finds copies of the same file. If a generated portrait has been published on a gallery or reused on other profiles, a search can surface that. A freshly generated face has no copies, so the search comes back empty. ### Why would a scammer use an AI-generated face? A stolen photo can be traced back to its owner with an image or face search. A generated face has no owner and no past, so those searches have less to find. That is why an empty search on a profile that claims a long, public life deserves a closer look.