How can you tell if a picture is AI-generated?
You tell if a picture is AI-generated by stacking independent clues, not by finding one magic sign. Knowing how to tell if an image is AI generated is less about a sharp eye and more about a method: look, read the file, trace the source, then test it.
Each method catches different fakes and misses others. The table shows what each one can and cannot tell you, so you know which to reach for first.
| Method | What it can tell you | Main limit |
|---|---|---|
| Visual inspection | Anatomy, text, light and texture errors typical of generators | Newer models make fewer visible mistakes |
| Metadata and Content Credentials | The tool that created or edited the file, sometimes signed | Usually stripped when an image is re-uploaded |
| Reverse image search | Earlier copies, the original source, the first upload date | A brand-new image has no history to find |
| AI image detector | Statistical traces of generation that the eye cannot see | Can be wrong, especially on small or compressed files |
What visual tells give away an AI image?
The most reliable visual tells are details that require real-world logic: anatomy, readable text, consistent light and objects that interact. Image generators assemble plausible patches of pixels, and they still slip where several parts of a scene must agree with each other.
Zoom in to full resolution and inspect the image area by area. A thumbnail hides almost everything.
Hands, fingers and teeth
Count fingers and joints, then look where hands touch objects. Fused fingers, a thumb on the wrong side, a hand that melts into a cup or a ring that passes through skin are classic tells. Teeth can show the same problem: a continuous white strip or too many incisors.
Text, logos and signs
Read every word in the frame, including signs in the background, book spines, badges and labels. Generators are much better at short headlines than they used to be, but small text often turns into near-letters that look like a language you almost know.
Light, shadows and reflections
Find the main light source and check that every shadow falls away from it. Then check mirrors, windows, glasses and eyes: a reflection should show the same scene from the right angle. Catchlights that differ in shape between the two eyes are a subtle but useful clue.
Backgrounds and edges
Backgrounds get less of the model's attention. Look for railings that stop mid-air, windows of different sizes on the same facade, crowds whose faces dissolve, and hair or fur that blurs into the backdrop instead of ending in strands.
Skin, texture, symmetry and jewelry
- Skin that is airbrushed everywhere, with no pores, fine hairs or uneven tone, on a photo that claims to be unedited.
- Repeating textures: identical bricks, leaves or fabric patterns that tile across the image.
- Earrings that do not match, glasses with frames that change shape, or a necklace whose chain merges into the collar.
- Faces and buildings that are a little too symmetrical, or symmetry that breaks in one odd spot.
Why are the visual tells disappearing?
Visual tells are disappearing because every new generation of image models is trained to fix exactly the errors people learned to spot. Hands, readable text and lighting have all improved, and a photorealistic portrait can now pass a careful look.
Editing makes it harder still. A creator can generate many versions, keep the cleanest, retouch the hands and add film grain. Treat a missing tell as neutral, never as evidence that the picture is real.
Tells survive longest in complex scenes. Groups of people touching each other, several hands in one frame, crowded street signs, musical instruments and machinery all require many parts to agree. A simple portrait against a blurred background gives the model the least chance to slip, and gives you the least to inspect.
How do you check if an image is AI generated with metadata and Content Credentials?
You check the file's metadata because it can name the tool that made the image. This is the fastest test when you have the original file, and the least useful when you only have a screenshot.
- 01Open the file properties or an EXIF viewer and look at the camera model, software and creation date fields. A real camera photo usually lists a make, model, lens and exposure settings.
- 02Look at XMP fields for a digital source type. The IPTC vocabulary uses the value “trainedAlgorithmicMedia” for media created with generative AI.
- 03Check for Content Credentials, the C2PA standard that attaches a signed history to a file. Some generators and editing tools, including Adobe Firefly, add them automatically.
- 04If the history names an AI generator, you have strong evidence. If the fields are empty, you have learned nothing either way.
Social networks and messaging apps usually strip metadata when you upload, so a viral picture rarely carries any. Our guide to what C2PA and Content Credentials are explains how the signature works and why it can vanish, and the Content Credentials checker reads it for you.
Does reverse image search detect AI images?
Reverse image search does not detect AI images directly, but it often answers the real question: where did this picture come from? Upload the image, or a crop of its most distinctive part, to two or three reverse search engines and sort the results by date.
- An earlier copy on an AI art community or a prompt-sharing site settles the matter.
- A copy credited to a named photographer or a news agency points the other way, and you can contact them.
- Slightly different versions of the same scene, with changed faces or hands, suggest several generations from one prompt.
Two habits improve your results. Search a tight crop of the face or the main subject as well as the whole frame, since reposts are often cropped or framed differently. And try a mirrored copy, because some reposts flip the image to dodge duplicate filters.
The limit is simple: a freshly generated image has no history to find. No match is not a sign of authenticity. It only means you need the other checks.
How do you check the source and context of an image?
You check the source by asking who posted the image first and whether the scene could have happened as shown. Context checks catch fakes that look perfect, because a generator can invent pixels but not a consistent paper trail.
- Account: is it new, anonymous, or posting many similar dramatic images?
- Event: do other people, angles or outlets show the same moment?
- Place and time: do the weather, signage, season and architecture match the claimed location and date?
- Motive: does the image push you to share, donate, invest or feel outraged right now?
These checks matter most for profile photos and breaking news. If you are vetting a match or a seller, see how to handle fake dating profiles. Newsrooms will find a fuller workflow in our page for journalists and fact-checkers.
How do you use an AI image checker and read the result?
You use an AI image checker as one more witness, and you read its result together with what you already found. A detector sees statistical traces the eye cannot, but compression, cropping and heavy editing weaken those traces.
Upload the largest, earliest copy you can find rather than a screenshot. The AnyFact AI image detector accepts JPG, PNG, WebP and HEIC files up to 20 MB, or a link. It checks provenance first, then runs an open-source AI-image classifier, then lists visible anomalies such as hands, lettering, shadows, reflections and textures.
The verdict comes in three levels: probably human, uncertain or probably AI. It always shows the signals behind it, a plain-language explanation and the limits of the method. Read “uncertain” literally and lean on the other checks. Our guide on whether AI detectors are accurate explains why no score should be read as proof.
AI Photo Detector · Free plan
Check an image for AI generation: provenance, an open-source classifier and listed anomalies, in one report.
Which images are hardest to check?
The hardest images to check are the ones that mix real and generated content, or that went through so many apps that every trace is blurred. Knowing these cases in advance stops you from reading too much into any single result.
| Case | Why it is hard | What to do |
|---|---|---|
| Real photo with an AI edit | Most of the pixels are genuine; only an object, a sky or a face was added or removed | Look for the edited area: a sharper or softer patch, lighting that does not match, a missing shadow |
| Beauty filters and phone processing | Smoothing and sharpening mimic the plastic look of generated skin | Ask for the unfiltered original or a second photo from the same moment |
| Illustrations and digital art | Human artists also produce clean, stylized images with impossible light | Check the artist’s portfolio, sketches and process posts |
| Screenshots and memes | Metadata is gone and compression hides fine detail | Trace the image back to its first upload before judging it |
| Small or cropped images | Too few pixels for a detector or a human to see much | Find a larger version with a reverse image search |
Art raises its own questions about style and process; our AI art detector page covers that use. For edited photos, remember that “partly AI” is a real answer: a genuine scene with a generated background is still a manipulated image.
How to tell if an image is AI generated: a step-by-step checklist
Run this AI image test in order and stop when you have enough evidence for the decision you need to make. The early steps are the cheapest, and they often settle the question before you need to study pixels.
- 01Find the highest-resolution, earliest copy of the image.
- 02Read the metadata and Content Credentials. Note any generator named.
- 03Run a reverse image search and sort results by date.
- 04Check the source account and whether other evidence shows the same event.
- 05Zoom in on hands, text, light, reflections, backgrounds, skin and jewelry.
- 06Run the image through an AI image detector and read the signals, not just the verdict.
- 07Weigh everything together. If the checks disagree, say “unverified” rather than guess.
Write down what each step found, even when it found nothing. A short note such as “no metadata, earliest copy from an anonymous account two days ago, hands look wrong, detector says probably AI” is far more convincing to an editor, a platform or a colleague than a single score.
The same logic applies to moving pictures, with extra clues in motion and sound: see how to spot a deepfake video.
AI Video & Deepfake Detector · Free plan
Got a video instead? Score it frame by frame and see suspicious moments on a timeline.
Frequently asked questions
Can you tell if an image is AI generated just by looking at it?
Sometimes. Errors in hands, text, shadows and reflections are still common, but recent generators avoid many of them. A clean-looking image is not proof of authenticity, so combine your visual check with metadata, a reverse search and a detector.
Is there a free AI image checker?
AnyFact has a free plan. a free account gets 10 credits a month with no credit card, and a photo analysis costs 1 credit. You can also run one photo analysis from the home page before creating an account; the result is shown once you sign up.
Does a screenshot remove signs of AI generation?
A screenshot removes the metadata and Content Credentials and adds compression, which makes detection harder. Visual tells and the source trail are still usable, so look for the original file whenever you can.
Do AI images always have a watermark?
No. Some generators add a visible logo, an invisible watermark or Content Credentials, but many do not, and visible marks are easy to crop. The absence of a watermark tells you nothing about how an image was made.
Can an AI image detector be wrong?
Yes, in both directions. Heavily edited, filtered, compressed or very small images are the hardest cases. Treat any detector result as an indication, and never sanction a person on a detector result alone.
What should I do if a picture of me was faked?
Save the image, the link and the account name, then report it to the platform using its impersonation or synthetic media reporting option. A detector report with the signals listed can support your report, though it is not proof on its own.