4 Simple Ways to Identify AI Generated Images (2024)

It seems hard to believe that AI-generated images became available to the public less than a year ago. They’ve already taken over all relevant visual mediums, from social media and artistic expression to marketing and image licensing, in a matter of months.

But this flood of AI-produced pictures creates a new problem: telling a “traditionally made” image from an AI-generated one.

This is something you might want to be able to do since AI-generated images can sometimes fool so many people into believing fake news or facts and are still in murky waters related to copyright and other legal issues, for example.

For that, today we tell you the simplest and most effective ways to identify AI generated images online, so you know exactly what kind of photo you are using and how you can use it safely.

Let’s go!

What Are AI-Generated Images?

AI-generated images are those created by artificial intelligence applications, namely, AI generative models based on GAN (Generative Adversarial Networks) technology.

If that sounded like gibberish to you, know this: AI image generators use intelligent algorithms that can correlate data –text descriptions and image content that the software already knows– to produce a completely new image based on a text prompt introduced by the (human) user.
AI video generation works similarly; here, you can read all about that.

Even simpler? They’re tools where you can create images by writing a description of what you want, and the software makes the image for you.Some tools, like Mokker AI, don't even need you to type in instructions, you can use preset buttons to define the type of image you want, and it creates it (in the case of Mokker, it's product photos).

Additionally, there are AI-generated images known as “deepfakes.” Deepfake technology mixes artificial intelligence with facial recognition software to manipulate videos or photos of real people –for example, celebrities– into realistic-looking images. This type of AI imagery is a bit more problematic, as you will soon learn.

AI-generated images have become a trend in recent times –a big one if you go by these latest visual AI stats— because they provide an alternative to the laborious task of manual image creation. At the same time, they expand the creative possibilities of the visual art design. The tech that makes them possible keeps improving quickly, resulting in very realistic and visually impressive AI-generated pictures that could easily fool the unsuspicious eye.

Why it Is Important to Identify AI Generated Images (and How to Do It)

As they become more common and more present in digital mediums, even in stock image libraries available for license, it is increasingly important to identify whether an image is AI-generated, mainly because their legal background is different from traditional, human-made pictures and because sometimes they might confuse viewers (like in the case of deep fakes).

As they’re so new, there is no universally-accepted standard for copyrighting AI-generated images. Still, the incipient legal frame points out that they are not copyrightable.

When users use AI generative tech to manipulate the image of a real-life person –putting other faces in someone else’s body and such– the result can be realistic and believable. And even if the creator clarifies that it’s an AI-generated picture, those important details are commonly lost if it gets shared around –like on social media. If you’re not careful, you might fall for misinformation and fake events, like recently with the fake photos of Donald Trump being arrested or Pope Francis wearing a designer jacket.

Furthermore, many people are questioning the legality of synthetic media, as they’re technically built from “bits” of other (human) artists’ work, often without authorization or compensation. Some are even suing AI generative app developers for copyright infringement.

This means that, as of right now, no AI generative tool can guarantee the legal validity of the images created with it… and that neither you nor they own the copyright of said images. While AI-generated photos are fun to create and visually impactful, some AI image generators are safer and ethically fairer than others; all of this makes it necessary to know when an image you see is GAN generated image and what rights the generator or the licensing agency is giving you over them if you decide to use it.

Now you know why it’s so important, let’s see the ways in which you can easily tell when an image is AI-generated.

Because there is still so much being questioned regarding AI-made visuals, companies that generate them and/or that license them do everything they can to be transparent about their origin.

A big part of that involves the image’s metadata. Stock photo agencies that admit AI images in their libraries demand that contributors label the files as AI-generated; in the image title, description, and image tags (which makes it easier to search for or exclude AI-created photos when surfing their catalogs). Looking for these labels is the simplest way to spot an AI-generated image.

Note: When images are being shared on social media, websites, and or digital platforms, they also tend to disclose the nature of the pictures (mostly to avoid any legal repercussions). It might be included in the photo caption or an adjacent note. Why do we mention this? Because it’s fairly common to skip reading these bits of text when the image is too captivating or interesting, and because you can also count on some users intentionally “burying” the disclosure or making it less obvious. So it’s important to always read captions and notes before assuming an image is traditionally made and veridic.
Additionally, checking who has shared the image on social media can help you determine its authenticity. For example, in the case of the latest (and fake) image of Julian Assange, an author pointed out how the fact no one close to Assange had shared the image was one extra way to discredit its veracity.

#2. Find the Watermark

Many AI image-generating apps available today issue watermarks on the images created with them, especially if they are done with a free-of-charge account. Not all are prominent, but you can always watch out for a small company logo –which means you’ll have to verify if the brand belongs to an AI image generator– or text indicating that the image was produced using AI tech.

Some others are less evident; Dall-E, for example, watermarks images downloaded from its platform with a string of five colored squares at the bottom right corner. So you need to know what to look for.

Again, super simple and “old school,” but effective nonetheless.

#3. Look for Distortion or Anomalies

A foolproof way to spot an AI-generated image is to look for anomalies: visual errors caused by the imperfect functioning of the machine learning algorithms during the creation process. Uneven eyes, unnatural-looking teeth, missing or misshaped body parts –AI generators seem to have a hard time with human hands, for example–, glasses that merge onto the person’s face, cats with a tail in the wrong place, etc.

Other visual distortions may not be immediately obvious, so you must look closely. Missing or mismatched earrings on a person in the photo, a blurred background where there shouldn’t be, blurs that do not appear intentional, incorrect shadows and lighting, etc.

If you find any of these in an image, you are most likely looking at an AI-generated picture.

However, AI generative models –like Midjourney, Stable Diffusion, or Dall E 2– seem to release an improved version of their apps by the day, each time producing better quality imagery. Hence, it’s still possible that a decent-looking image with no visual mistakes is AI-produced.

This method is the more sophisticated and the only automated one. However, we list it last because the applications that promise to detect AI generation are not entirely accurate. Not yet, anyway.

There are apps designed to flag fake images of people, such as the one from V7 labs. But while they claim a high level of accuracy, our tests have not been as satisfactory.

Microsoft has its own deepfake detector for video, the Microsoft Video Authenticator, launched back in 2020, but sadly it’s not entirely reliable when it comes to spotting AI-generated videos.

Some companies are developing GAN detector software specifically designed to spot AI-generated images. Mayachitra’s GAN detector is one said tool where you can upload an image to be analyzed and told whether it’s AI-generated. However, sometimes it simply fails.

This doesn’t mean these apps aren’t useful, just that you must use them in connection with other strategies –such as the previous 3 methods described– to ensure precise results.

Plus, you can expect that as AI-generated media keeps spreading, these detectors will also improve their effectiveness in the future.

FAQs in Relation to Identifying AI Generated Images

Spot the AI in Your Images

And there you have it. These four easy ways to identify AI generated images will help you be always certain of the origin of the content you use in your designs and, equally important, the content you see and consume online.

Which one do you think it’s the simplest and most effective? Go to our comments section and let us know!

Image header: Copyright by davidpereiras / photocase.com, all rights reserved.