Guide
How to Tell if an Image Is AI Generated: 7 Checks
How to tell if an image is AI generated, step by step. Seven checks you can run yourself, what each one can and cannot prove, and when to use a detector.
Last updated Nov 14, 2026
In 30 seconds: no single check proves an image was made by AI. Look closely at details, light and backgrounds, then check the file's metadata and Content Credentials, search for the image's origin, and confirm who first posted it. Use an AI image detector last, as one more piece of evidence, not as the verdict.
| Check | What it can show | What it cannot show |
|---|---|---|
| 1. Details | Typical generator mistakes | A clean image is not proof of a real photo |
| 2. Light and shadows | Scenes that could not exist | Editing can cause the same problems |
| 3. Backgrounds and edges | Melting or repeating patterns | Compression can blur real photos too |
| 4. Metadata and Content Credentials | A tool's own label on the file | Missing data proves nothing |
| 5. Reverse image search | Where and when the image appeared | New images have no history yet |
| 6. The source | Whether the poster is credible | A real account can share fakes |
| 7. A detector | A model's estimate | Certainty either way |
1. Look closely at the details
The fastest way to tell if a photo is AI generated is to zoom in on the parts generators still get wrong. Look at hands, teeth, ears, jewelry, glasses and any writing in the scene.
- Hands and fingers: extra or merged fingers, nails in odd places, hands that blend into objects.
- Text: signs, labels and book covers with letters that look almost right but spell nothing.
- Symmetry: earrings that do not match, glasses with different frames on each side, buttons that change shape.
Newer generators make fewer of these mistakes, so a clean image is not proof that it is real. Treat a clear mistake as a strong sign and the absence of mistakes as no sign at all.
2. Check the light, shadows and reflections
Real scenes follow physics; generated scenes only imitate it. Check whether every shadow falls away from the same light source, and whether reflections in windows, water, mirrors and eyes match what is in front of them.
Common giveaways are shadows pointing in different directions, a face lit from the left in a room lit from the right, or a reflection that shows a different scene. Heavy photo editing and composites can cause the same problems, so these signs show that an image was altered, not necessarily that AI made it.
3. Look at the background and the edges
Generators spend their detail on the subject and often lose control of the background. Look for faces in a crowd that melt together, railings and tiles that repeat or bend, and objects that merge into each other.
Check the edges of hair, fur and clothing against the background as well. A soft, painted look where the subject meets the background is common in generated images. Remember that heavy compression and upscaling can blur real photos too, so weigh this check alongside the others.
4. Read the metadata and Content Credentials
The file itself can carry a label. Many AI tools write their name into the image's metadata, and some add Content Credentials, a signed record based on the C2PA standard that says which tool made or edited the file.
- EXIF and XMP metadata: camera photos usually record the camera model, lens and settings. Some AI tools record the generator's name instead.
- Content Credentials (C2PA): images from several major generators, including OpenAI's, carry these records. A verification tool can read them and check the signature.
- Invisible watermarks: Google marks images from its own models with SynthID, which survives some editing but needs Google's tools to read.
Our free image metadata viewer shows these labels in your browser, and our guide to Content Credentials (C2PA) explains how to read them.
Missing metadata proves nothing. Social networks and messaging apps strip it on upload, and a screenshot never has it. A credential that is present tells you what a tool claims; check that the signature is valid before you rely on it.
5. Run a reverse image search
A reverse image search shows where else an image has appeared. Upload it to Google Lens or TinEye and look for the earliest copy, other versions with different crops or details, and pages that explain where it came from.
If a supposed news photo first appeared on an AI art forum, or the same "photographer" posts a new impossible scene every day, that is strong evidence. If the image is brand new, a search may find nothing, which tells you nothing either way.
6. Check who posted it first
The source is often more telling than the pixels. Ask who first shared the image, whether they were in a position to take it, and whether credible outlets or people at the scene confirm it.
Look at the account's history: a new account that only posts dramatic images, or one that has shared generated images before, deserves more doubt. Real accounts can also share fakes by mistake, so a trusted source raises confidence without settling the question.
7. Use an AI image detector, and read the result carefully
An AI image detector is a model trained on large sets of real and generated images. It estimates how closely your image resembles each group and returns a score. It is useful when your own checks are inconclusive, but it is an estimate, not proof.
- Upload the original file where you can. Screenshots, heavy compression and resizing remove the detail detectors rely on.
- A result in the middle of the scale is a real answer: the evidence is mixed.
- An edited photo with an AI-filled area, or an AI image retouched by hand, can land anywhere on the scale.
Our AI image detector shows the score next to the file's Content Credentials and EXIF evidence, so you can see what the result was based on. The sample image report shows every part of a report, and how AI detectors work explains the scale.
When the checks disagree
Most real cases end with mixed evidence: an image looks clean, has no metadata and has no history. Then the honest answer is that you cannot tell yet. Ask the person who shared it for the original file, look for other photos of the same moment, and wait for confirmation before you share or act on it.
Never use a single check, or a detector score on its own, to accuse someone of faking an image. See AI detector accuracy and limits for where image detection fails.
Frequently asked questions
Can you always tell if a picture is AI? No. The best generators produce images that pass every visual check. Combining the source, the file and a detector gives you better odds, not certainty.
Does a photo without EXIF data mean it is AI? No. Most platforms remove metadata on upload, and screenshots never had it.
Is there a free way to check if an image is AI? The visual checks, metadata viewers, Content Credentials verification and reverse image search are all free. Our image check uses credits; new accounts get free credits when they sign up.