Real photo or AI photo?
AI generators now make photos that look like they came from a phone or a good camera: faces for profiles, houses for rental listings, products for shops, crowds and storms for news posts. This page is the detector set up for photos. Drop one or several photos on the light box above and each gets a reading, from human made (a camera photo) to AI made, with how sure the detector is.
The detector here is the same open model as on the home page, running in your browser. For photos, the Colour boost lamp is on when you look closer, because the colour noise of a camera sensor and the smoother, blotchier colour of many generated photos look different under it.
How it did on photos
In our test the detector cleared 95% of 59 real photos (featured pictures from Wikimedia Commons: people, places, plants, objects and animals), and caught 86% of 35 photo-like AI pictures (people, animals, food, places and objects made with named generators). The full list, including the photos it got wrong, is on How we test.
Where photo checks go wrong
- Phone processing. Phones smooth, sharpen and merge several frames into one photo. Night mode and portrait mode in particular can push a real photo toward AI made.
- Edited real photos. Heavy noise removal, beauty filters and AI upscaling change the fine detail the detector reads. A photographer noticed exactly this with real photos that were denoised in professional software.
- Re-saved copies. Every time a photo is posted, screenshotted and posted again, it loses detail. A fifth generation copy reads less clearly than the original.
- Real photos with an AI part. A real photo with a replaced sky or a removed person may still read as human made, because most of it is.
Checks to do with your own eyes
Use the loupe on the parts generators find hard in photos: hands and fingers, teeth, ears and earrings, eyeglasses, text on signs and shirts, the edges of hair against the background, reflections in windows and eyes, and objects that should be the same shape twice (two wheels, two shoes). In real photos, background people have faces and fences are straight; in many AI photos, background detail turns to mush on a closer look.
Also look outside the picture: who posted it first, whether other photos of the same moment exist from other angles, and whether a reverse image search finds an older copy. A photo of a real event usually has more than one witness.
Profile photos and listings
Fake profiles and fake listings often use AI faces and rooms because they are free and look new. If the detector reads a dating or seller profile photo as AI made, look at the other photos on the profile too: AI profiles often have only a few photos, all with the same flawless light. A reading alone is not proof that a person is fake,
Shops, rentals and news photos
Generated photos also turn up as product shots for items that do not exist, rooms in rentals that look nothing like the real flat, and dramatic weather or crowd photos shared as news. In each case the cost of believing a fake is real money or a wrong idea of what happened. Check the photo here, then ask for something a generator cannot easily make: a short video of the item turning in someone's hand, a photo of the room with today's newspaper, or the same scene from a second source. A real seller or a real witness can usually give you that within minutes.
Questions
Can it tell a phone photo from an AI photo?
Most of the time: on our test it cleared 95% of real photos and caught 86% of AI photos. Heavy phone processing and filters can still mislead it.
Does a filter make my real photo look AI?
It can. Beauty filters, AI upscalers and strong noise removal change the fine detail the detector reads, and can push the reading toward AI made.