how to tell if an image is ai generated how to tell if a picture is ai or real how to check if an image is ai generated free

How to tell if an image is AI generated: 5 checks

Adrian Maverick · · 9 min read

You usually cannot prove an image is AI generated just by looking at it, but you can get close by stacking a few checks. Look for the visual slips, then check for a watermark or a Content Credentials label, then find out where the image first appeared online.

No single check is final. A clean result from one tool only rules out what that tool can see, and the people who build these tools say so on their own help pages.

Look for the visual clues, and know where they fail

Both Google and the FBI publish lists of what to look for. The FBI's December 2024 warning about AI in fraud lists subtle imperfections such as:

  • distorted hands or feet
  • unrealistic teeth or eyes
  • indistinct or irregular faces
  • unrealistic accessories such as glasses or jewelry
  • inaccurate shadows
  • watermarks

Google's Gemini help page on verifying AI content adds a few more: inconsistent or garbled text inside the image, errors in small background details, inconsistent lighting, and repetitive patterns that seem unnatural.

These clues are worth a look, especially in the background and in any lettering on signs, labels or badges. But they catch the sloppy images, not the careful ones. The FBI's own warning opens by saying it "can be difficult to identify when content is AI-generated", and a clean, well-made fake will pass a visual check. Treat a visual slip as a reason to dig further, and treat the absence of one as no answer at all.

Check for a Google SynthID watermark

SynthID is Google's invisible watermark for AI content. According to Google DeepMind, it is added the moment content is created by Google's AI and is designed to survive cropping, filters and lossy compression.

To check an image in Gemini, Google's help page gives these steps on a computer:

  1. Sign in at gemini.google.com. You need a personal account or a qualified Workspace account.
  2. Click Add files, then Upload Files, and pick the image. One file at a time, 100 MB or less.
  3. Ask a question such as "Was this image created or edited by Google AI?" or "Is this AI-generated?"

If you only have a screenshot, Google says to crop tight around the image and not to upload a collage of several images.

Read the result carefully. Google's help page says:

  • If a watermark is found, all or part of the image was created or edited by Google's AI.
  • If no watermark is found, the image "wasn't created or edited by Google AI, but it could have been created by other AI systems."
  • Some results are unclear, with reasons such as "Not enough details to watermark" or "Likely too small an edit".

The same page says Gemini can currently only recognize content created by Google AI tools, and that the watermark may not be detected after many alterations. There is also a usage quota of approximately 10 image checks in any rolling 24 hours.

Google DeepMind also runs a separate portal, SynthID Detector, which it says anyone can use to check whether an image, video or audio file "was made with AI from Google or our partners, including OpenAI, NVIDIA, Kakao, and soon, Apple." So a "no watermark" answer still leaves every other AI image tool unaccounted for.

Look for Content Credentials

Content Credentials are a different idea. Instead of a hidden watermark, the file carries a signed record of where it came from and how it was edited. The standard comes from the Coalition for Content Provenance and Authenticity (C2PA), which describes it as working "like a nutrition label for digital content".

Where you can see them:

  • In Gemini. The same verification check reads Content Credentials and can show a summary of the media's composition, its edit history and whether AI was used, along with the company that signed the information. Google's help page says this works at gemini.google.com and in the Gemini app for Android, with iOS coming.
  • In Google's About this image. Google said in a September 2024 post that if an image contains C2PA metadata, About this image can show whether it was created or edited with AI tools.
  • On the Content Credentials site. The C2PA's contentcredentials.org site links to a verify page where you can inspect a file.

What they cannot do matters just as much. Google's help page is direct: Content Credentials do not tell you if a file is definitely AI generated. They tell you what the signing app or device recorded. The C2PA's own FAQ says credentials can be separated from the file, and NIST's report on synthetic content notes metadata is stripped "as it often is when files are shared (e.g., via social media platforms)."

So a valid credential is useful information. A missing one tells you almost nothing, because most images online, real or fake, do not carry one.

Use About this image and reverse image search

Often the fastest route to the truth is finding where the picture came from. A real news photo usually has a trail back to a named photographer or outlet. An AI image made for a scam usually appears out of nowhere.

About this image. In Google's 2023 announcement, the tool shows when the image and similar images were first indexed by Google, where it may have first appeared, and where else it has been seen online, including on news and fact-checking sites. Google says it is reachable from Google Images, Lens and Circle to Search. If fact-checkers have already written about the image, this is where you will see it.

Reverse image search with Google Lens. Google's help page gives three ways on a computer:

  1. Go to google.com, click Search by image, then Upload a file.
  2. Or copy the image's address, click Search by image and paste it into "Paste image link".
  3. Or in Chrome, right-click the image and choose Search with Google Lens.

Results include websites with the image or a similar one. Look for the earliest and most credible source, and for the same image with a different caption, which is a common sign of a recycled or edited picture.

Check the metadata, if you have the original file

If someone sent you the original file rather than a screenshot, its metadata can help. Google's help page points out that photos can carry EXIF data with the camera model, dates and settings, and that metadata sometimes shows traces of editing software.

Two limits apply. Metadata is easy to remove and easy to edit, as NIST notes, and messaging apps and social networks often strip it on upload. Missing EXIF data is normal and proves nothing either way.

Why AI detector websites cannot settle it

It is tempting to paste the image into a free "AI detector" and take its score as the answer. NIST's 2024 report on synthetic content explains why that is risky. Detectors trained on one image generator do worse on images from others, and after compression and resizing the error rates get large: one 2023 paper it cites found accuracy of 61% to 70%, and a 2022 paper found 50% to 62%.

A score from a detector is one more signal, not a verdict. If the stakes are real, rely on the source of the image and on checks you can verify.

When an image is being used to get something from you

Most of the time the question is not "is this AI?" but "should I trust what this image is asking me to believe?" The FBI warns that criminals use AI images for fake social media profiles, fake IDs, photos sent to "prove" an online contact is a real person, and pictures of disasters to raise money for fake charities.

If an image is part of a request for money, personal details or a login, the image is not the thing to verify. Verify the person or organization through a channel you already trust, and do not send money to someone you have only met online. If you have lost money, the FBI asks you to report it at ic3.gov. For the same checks applied to video and live calls, see how to tell if a video is a deepfake, and for voices, how to tell if a call is AI.

Frequently asked questions

Can you tell if an image is AI generated for free?

Yes, with limits. Google Lens reverse image search is free, and Gemini's verification check is available to signed-in users with a daily quota of about 10 image checks. Neither can confirm every AI image, so combine them with the visual clues and the image's source.

Does a missing SynthID watermark mean the image is real?

No. Google's help page says a missing watermark means the image was not made or edited by Google AI, but it could have been created by other AI systems.

How do I tell if a picture is AI edited rather than fully generated?

SynthID can flag an image that was edited by Google's AI, though Google says very small edits may not carry a detectable watermark. Content Credentials, where present, can list the edit history. Reverse image search can turn up the unedited original.

Can metadata show if an image is AI generated?

Sometimes. Some AI tools and cameras write Content Credentials or other metadata into the file. But metadata is often stripped when files are shared on social media, so its absence proves nothing.

Are the hands and fingers still a giveaway?

They can be, and the FBI still lists distorted hands as a sign. Newer tools make fewer of these mistakes, so a normal looking hand is not proof the image is real.

Keep scam images out of your inbox

Fake images often arrive by email, attached to an invoice, a profile, or a story meant to rush you. Our guide on how to spot AI phishing emails covers the wider pattern.

One habit helps: give each site its own address. In Secria Mail, aliases make that easy, so if a targeted message lands on an alias you only gave one shop, you know where it came from and can delete the alias to stop it. Tracking pixels are removed automatically and senders never see your IP address. Sign-up is a name, the address you want and a password, with no phone number. Create a free account.

Secria fact-checks every post against primary sources. Spotted something wrong or out of date? Email hq@secria.me and we will correct it.