April 27, 2024, 6:40 am | Read time: 6 minutes

Is this image real or fake? Internet users increasingly have to ask themselves this question. It can involve seemingly “harmless” cases like images of Princess Kate and her children. However, it can also raise general questions of credibility, such as in the case of seemingly Netflix-edited images in a true-crime documentary. AI-edited images are also increasingly used for disinformation; you can no longer rely on what you see. But how can you possibly tell if an image has been edited by AI?

AI-edited images are already widespread

In principle, it is important to distinguish whether an image was completely created using AI, such as with programs like Midjourney. Even long-established editing software like Photoshop has long relied on artificial intelligence to create certain effects.

Partial image editing is a well-known and accepted method in professional photography. Retouching a wrinkled clothing fold or skin blemishes is not only common in social networks. However, it becomes problematic when content is altered and presented as reality.

An example: You see an image on social networks of a well-known politician in an unflattering situation, such as shaking hands with a dubious-looking person. Such images are increasingly generated to harm political opponents. Another recurring example: fake nude images of major stars. For instance, Taylor Swift was a victim of pornographic deepfakes. And it’s not just celebrities who are affected. With the right software, such as in so-called “Nudify” apps, any image on the internet can be altered to show the person in a completely fabricated situation.

Problems with AI-generated images

This is especially problematic when such content is not labeled as such. Publishers are actually required–just like in the text area–to always indicate that AI was involved in the creation. However, this does not always happen. And private individuals who edit an image with such a program do not always make this clear. In the worst case, the personal rights of those affected are violated, leading to severe defamation.

The BSI (Federal Office for Information Security) describes several problematic scenarios in this context:

• Overcoming security procedures
• Social engineering
• Disinformation campaigns
• Defamation

To recognize whether an image is real and AI-edited, it is worth paying attention to a few points on your own.

Recognizing AI images with some clues

A good starting point is always the hands and fingers of a depicted person. Often, the posture in AI-generated images looks unnatural, and a finger may be missing–or there is suddenly one too many. Eyes are also difficult for AI to depict. Therefore, look closely at the light reflections and eyelashes. Too perfect results with completely symmetrical eyelashes are generally not real. Speaking of light: Always look closely at the lighting conditions in the image. Do the shadows match the light source? Is it generally realistic? Do the reflections on reflective surfaces match?

Further clues in human objects are often found in the person’s hair. Is the parting overly pronounced? Or does the structure seem strange to you? Often, upon closer inspection, you will also discover some hairs that seem to have taken on a life of their own and “float” in the image without connection to the head. A look at the mouth of the people can also be worthwhile. Often, for example, the transition from lips to teeth is somewhat blurred, or the teeth generally appear too exaggerated.

More on the topic

Growing Danger How to Check if Your Photo Is Being Used for Deepfakes

Personal Rights, Abuse, … Sharing Vacation Photos on Social Media: What to Keep in Mind

It’s all about the details

The clues mentioned above were exclusively about recognizing AI images that show human beings. This is somewhat easier for most people because we are used to seeing people in our daily lives. Inconsistent details are therefore more noticeable. On the image of an owl, however, fewer people would know which details might be unusual or even impossible.

In general, the same applies here: Keep an eye out for details. It is worth looking especially at areas that show particularly small objects, such as the feathers of the owl. In landscape photos, these can also be branches. Fine branch structures in AI-generated images can exhibit unnatural shapes and connections. Often, problems with contrast and clarity also occur in this context. Individual structures are suddenly slightly blurred, which would not be the case in a real photo.

Often, particularly delicate structures like leaves or blades of grass are also very repetitive. It may be that you see certain sections repeatedly. Lighting conditions are also a good clue in this case to recognize AI images. Do light rays really point to a common center? Does the shadow cast make sense in this context?

Currently, it is also the case that artificial intelligence generally has problems with the correct representation of text and other characters. These often appear somewhat blurred or make no sense at all. So if you see numbers or characters in the background of an image, zoom in if possible and examine the image section closely.

Further measures to recognize AI images

In addition to these clues that can be found in the images themselves, you can also do some research outside of that to verify the origin of an image. If you found the image on a social medium, take a look at the profile that posted it. Are there similar images there, and if so, in what quality?

If such an image shows a well-known person, you can also check the news. Do reputable media report on the events shown? It is also worth saving the image and then checking the metadata through the settings. These include the creation date and the author. AI programs like DALL-E also label their images accordingly. Another way to recognize AI images is the reverse search on Google. This allows you to find out who posted an image first and where else it can be found.

All of this, however, requires that you constantly critically engage with what you see. You can no longer rely on the fact that what is depicted truly reflects reality. Experts assume that the use of AI images will increase significantly. Therefore, it is important to pay attention to certain clues and always question the plausibility of an image and information.

An important note at this point: AI technology is currently developing very quickly. This also applies to image AI. Details that were obviously recognizable as fake a few months ago already appear much more realistic today. Therefore, the clues mentioned are no guarantee of directly recognizing an AI-generated image as such.