FREE
Loading…
Two pictures, one is a real photograph. Five questions a day, easy to hard — and up to 5 credits for a sharp eye.
Loading today’s round…
Step 1
Each question shows two pictures of the same kind of scene. One is a real photograph. The other was generated by an AI model.
Step 2
Tap the one you think a camera took. You get one try per question, and the answer is revealed right away along with the giveaway.
Step 3
Questions get harder as you go: two easy, two medium, and a hard one where the AI image recreates the very same scene as the real photo.
The six checks that decide most questions in this game.
Signs, labels and license plates are where generators still slip. Real text is legible and consistent; AI text often dissolves into near-letters.
Hands, teeth and ears hold up badly at the edges of the frame, where the model paid less attention.
Real light comes from somewhere. Look for shadows that point different ways, or reflections that show a different scene.
Evenly spaced objects, flawless skin and symmetrical clutter are typical of generated images. Real scenes are messier.
Crowds, shelves and foliage behind the subject tend to melt into repeating texture in AI images.
In hard questions the AI image was generated from the real photo itself, so place, light and framing match. Only small details — a sign, a hand, a flag’s folds — are left to judge.
Want the opposite — an AI image that looks like a real photo? Try the AI image humanizer.
Yes. Practice mode is unlimited and needs no account. The daily round of five questions needs a free sign-in so we can pay out credits.
Each right answer is +1 and each wrong one is −1. When you finish the round the total is added to your account, never below zero: 5 right earns 5 credits, 4 right earns 3, 3 right earns 1. You can earn at most 5 credits a day.
No. Wrong answers only lower that day’s winnings. Credits already in your account are never deducted by the game.
The real photos are public-domain and CC0 photographs from Flickr, found through Openverse, and uploaded before AI image generators existed. The AI images are generated by us with GPT Image 2 and Gemini, from a description of the same scene — or, in hard questions, from the real photo itself.
Every day at 00:00 UTC. Everyone plays the same five questions, which is how we can show real statistics like how many players got a question right.
Current generators are very good at texture, and in the hard questions the AI image is recreated from the real photo, so the scene matches almost exactly. What still gives them away is usually logic: text, hands, light and what a real place would actually contain.
Last updated: September 23, 2026