If you’ve been on the internet lately, it’s more likely than not that you have seen an AI generated image, even if you didn’t realize it. Posts, videos, and even album covers, all generated with AI. A lot of people are tired of it. A lot of people are vulnerable to it. But after reading this article, you might be a little more familiar with the tells of AI generated images. 

Here’s our first example of an AI image. Four different signs of generative AI have been circled in red. To the more trained eye, you can spot many more. The problem with trying to identify images like this is that they’re mimicking artists who make mistakes. It’s important to know the difference between AI mistakes and human mistakes so as not to call real art AI.

Here are the points of interest:

1. The signs contained garbled, unreadable text. While one may argue that an artist could’ve done that as a stylistic choice, you can see that some the letters are actual letters, others aren’t, and with the ones that are, they only spell gibberish.
2. This part goes from being a branch to being a wire, and around the start, part of it fades into the leaves and blends into it. No human would consciously make that mistake.
3. The cup seems to almost be floating, along with it being a color that stands out too much. This image was originally animated, and the steam is a sign that is, for some reason, in most of these subtly animated images. It’s never moving naturally.
4. Could you tell what this is supposed to be? A good thing to keep in mind when trying to identify AI images is intent. Why would the artist make a random rectangle that doesn’t lead to anything or add to the image at all?

Hopefully those signs pointed out a few things you haven’t thought of before.

Keep those in mind as we identify our next image.

Along with the last one, this image was taken from a Lo-Fi studying video. This genre of video has recently become extremely saturated with AI generated looping animations. A few things that they usually have in common are rain, coffee with steam coming off of it, and using copyrighted “nostalgic” characters like Snoopy. In this case, it’s a little hard to tell which character it’s supposed to be. It looks like a mix between Jiji from Kiki’s Delivery Service and Chococat.

Once again, here are the points of interest:

1. The rock on the side of the water suddenly turns into the bush. At first glance, it looks like it could be covered by the bush, but the colors fade into green, and it doesn’t look natural. 
2. The fur on the top of the head spikes up strangely, which doesn’t look right and isn’t a feature the character usually has. When the image is animated, it shifts strangely, which is another thing to keep track of.
3. Similarly to the rock, the leaf blends into the bush underneath it. It looks partially like it’s supposed to be the top of the bush, and partially like it’s a separate thing.
4. The reflection doesn’t make sense. It adds strange bits to the character that aren’t there, and completely leaves out the book.

Now, let’s try to identify a more “realistic” AI generated image.

This image comes from a stock image website, which some people find counterintuitive. When people want stock images, they usually want real ones. This image is pretty easy to identify as fake. A good thing to think about is the context- why would there be coral and plants in a bowl tank? There’s no filter, so it would be difficult to keep coral and plants alive. Even if the coral is meant to look plastic, the plants don’t.

As usual, here are the points of interest:

1. The tank is much too shiny. If you’ve ever had a fish tank, you know it gets dirty extremely easily. The water is too clean, and the glass is too crystalline. Even if you haven’t had a fish tank, this is way too shiny for anything real. 
2. Look out for things that don’t quite make sense. What is this? It’s clearly supposed to be a tank decoration, but you can’t quite make out what it is. When checking for AI, make sure you can identify everything in the image.
3. This looks like it’s supposed to be coral at the bottom, and yet, at the top, it folds like it’s a plant. 
4. The fish’s fin splits into two pieces, which doesn’t look natural. You could argue that the fin is naturally like that, but it looks wrong, and fish like goldfish don’t have split fins like that.

But can Liberty students correctly identify AI images?

On the left, is an AI generated image of a fish tank. On the right is a real fish tank.

“The one on the left is too perfect,” Melody Gagliano (9) said, “Also, beta fish can’t all be in the same tank together.”

“The one on the left- there’s artifacting on the fish and it’s way too vibrant. And the lighting doesn’t make sense,” Abigail Scherder (9) noticed.

Obviously, they’re both right. Most people can tell which is AI generated from things like the perfection of it, the lighting, artifacting, and even things like the water.

Now, can you tell which image is AI?