Quick answer for one picture
You saw a picture in a feed, a group chat or the news, and you are not sure it is real. The fastest way to ask "is this image AI?" here: take a screenshot, come back to this page and press Ctrl+V (Cmd+V on a Mac), or on a phone press Choose pictures and pick the screenshot. The detector reads it on your device and puts a tab on it, usually within 15 seconds.
Then tap the picture. You will see the reading on a scale from human made to AI made, a plain sentence about how sure it is, and any AI label hidden in the file. Remember that a screenshot carries no such labels, so the notes will usually say none was found.
How much to trust the answer
On our test of 203 pictures with a known origin, the detector was right 91% of the time. That is good, but it means some pictures in every hundred are read wrong. It missed 15 of 98 AI pictures and wrongly called 4 of 105 human made pictures AI. So:
- A strong reading (AI MADE or HUMAN MADE) is a good sign, not proof.
- A LEANS reading means the detector is unsure. Look closer and check the source.
- Before sharing a picture as fake, or as real, check where it came from first.
Five checks that take a minute
- Zoom in on the details. Use the loupe on hands, text, jewellery, teeth and the edges of hair. Generators still slip on these.
- Look at the background. Real places have readable signs and people with faces; generated backgrounds often blur into shapes that make no sense.
- Find the first post. A reverse image search often finds an older copy, the original artist, or a fact check.
- Look for other angles. Real events are photographed by many people. One perfect picture with no other copies is a warning sign.
- Read the account. Pages that post only dramatic pictures with no sources, or that sell something, deserve extra doubt.
Why screenshots are harder
A screenshot is a new picture of your screen: it is resized, it may include the app's buttons, and it is often squeezed again when you send it. All of this blurs the fine detail detectors read. For the best reading, crop the screenshot to just the picture, and use the original file when you can get it.
What the answer looks like
The tab on each picture uses one of four words: AI MADE, LEANS AI, LEANS HUMAN or HUMAN MADE. Under the picture, the reading puts a needle on a bar that runs from human made on the left to AI made on the right, and says in a sentence where the detector put it, as a percentage. A needle near either end means the detector is confident; a needle near the middle means it could not decide, which is a useful answer too: it tells you to rely on the other checks above.
When the answer matters a lot
If a picture could hurt someone, sway a vote, or cost money, do not rely on any single detector, this one included. Ask the person who posted it for the original, look for reporting by news outlets that checked it, and give the benefit of the doubt where a person's reputation is at stake. Detectors, including ours, were trained on the generators that existed when they were made, and new ones appear every month.
Questions
Can I paste a picture?
Yes. Copy a picture or take a screenshot, then press Ctrl+V (Cmd+V on a Mac) on this page, or press Paste a screenshot.
Why does it say no AI label was found?
Most apps and social sites remove the hidden labels when a picture is posted, and screenshots never carry them. A missing label says nothing either way.