Can you tell which of these images are AI generated?

Generative AI has come a long way in a short period of time. So much so that I’m willing to bet most people can’t tell the difference between a real image and an image generated by AI, at least at first glance. Don’t believe me? Take a look at the image pairs below and try to figure out which one is real and which one isn’t.

I created all the AI images myself with Gemini Advanced, which uses Google’s Imagen software. The real photos, on the other hand, were taken by my colleagues here at AA with the phones they were testing. However, I had to change their aspect ratio to 1:1 so that they are the same size as the images created by Gemini.

The photo samples are quite varied, showing buildings, graffiti, and people. I also included two night shots to make things even more challenging.

So take a look at the image pairs below and try to figure out which photo is real and which one is AI-generated, but do it before reading the text since that’s where the answer is revealed. Good luck.

Let’s start with a difficult one. The low lighting makes this one hard, hiding some of the issues generative AI can have when producing an image. Both show an illuminated Ferris wheel and a building or two in the background.

Neither of them is picture-perfect due to the low light, making it difficult to spot the real image from the fake one at first glance  — at least without detailed inspection. So, which one is real? It’s the one on the left, while the right one was made with Gemini.

Look at these two handsome people! One of them is my colleague Ryan, and the other one is AI-generated. It’s hard to tell which is which at first glance, but upon closer inspection, one does look more “edited” than the other.

In my opinion, Imagen 3’s problem is that many of the images it creates, especially of people, look too picture-perfect instead of a bit more casual. In other words, they look edited. So while the real image, which is the right one, does contain a blurred background to make it look nicer and to highlight the subject, the AI-generated one takes things up a notch, especially with the shadow across the face.

I wouldn’t be able to tell the real from the AI image at first glance if I didn’t already know which was which. They both look realistic, especially because of all the details. You can see them on the wall in both images, with one being slightly damaged or dented in a few places and the other showing details and shadows between the rocks on the wall.

The image on the left is AI-generated, while the right one is real. It’s fascinating how much detail AI can add to an image, as it contains various graffiti as well as signatures from artists. However, a trained eye will notice that a lot of the text on the wall is pure gibberish, which gives it away.

Back to night shots that will make you ponder. Both images are from Paris and show the Eiffel Tower, but one only shows the bottom part, while the other highlights the upper part.

I thought the difference between them would be evident before I created the AI image, mainly because of all the details the Eiffel Tower contains. But at first glance, they both look real, making it hard to find the fake. However, a closer look reveals that the left image has almost perfect lighting and looks a bit more professional than the other one, especially when you look at the contrast between the tower and the dark sky. Perfect lighting can sometimes be a giveaway for AI images, just like the glitched text I mentioned in the previous section.

These two could have fooled me. One does look more professional than the other, but then again, gorgeous close-ups of flowers and plants in general with a blurred background can easily be created with a good phone like the Pixel 9 Pro XL, especially in good lighting.

However, the professional-looking one on the left is fake. I’d argue that there’s a bit too much detail in the photo, and the bokeh effect — the blurred background — is exaggerated. The one on the right looks a bit more casual, which is the type of image AI generally struggles with.

These two buildings are completely different. One has a more industrial look with exposed brick and metal rails on the balconies, while the other is a bit more classic and elegant — basically an old building that was renovated.

Both contain plenty of details, which would make you assume that it should be easy to spot the fake since AI can get details wrong frequently. However, spotting the AI image is hard unless you have a trained eye. The only thing I think gives it away are the balconies, which all have what appear to be multiple tables on them and a bunch of other objects — which is out of the ordinary for apartment buildings. There are also too many glass doors on each balcony, all of which are very close together, which doesn’t relate to the average floor plan of an apartment. And there’s also another one to the side, which doesn’t contain a railing, which a door/window of that size always does. So yes, the right image is the fake one, although you really need to look and think to figure it out.

Let me know how many of these you got right in the poll below. I’m curious to see how hard it is for people to distinguish real images from fake ones, although I do think most got at least some of these correct since not every image pair was super difficult.