Images generated by artificial intelligence, here are the methods to recognize them

Artificial intelligence (AI) makes it increasingly difficult to distinguish real images from images created artificially with ad hoc tools. However it is possible understand if an image is generated with AI by paying attention to some details. Among these there are errors in the representation of some anatomical partsone on all the fingers (but also the reproduction of the hair and skin of the characters portrayed in the photos), errors in the reproduction of objectsand the presence of strange artifacts.

What are AI-generated images?

For images generated with artificial intelligence we are referring to all those that are not produced with the aid of a camera, but which are created using AI software based on algorithms machine learning (machine learning) or deep learning (deep learning).

Among these applications text-to-image (i.e. which allow you to generate images starting from textual commands) there are, for example, GIVE HER by OpenAI (the same company that develops ChatGPT and Sora) and Midjourneyjust to name a few of the best known and most used.

These tools have been trained by the respective companies that developed them by feeding them a stratospheric quantity of images with which they managed to “learn” how to reproduce human beings who do not exist, everyday objects, realistic settings and so on.

How to understand if an image is made with artificial intelligence?

At the moment, artificial intelligence is not infallible, it very often generates images containing errors – more or less evident – which can provide useful clues to understand whether we are dealing with a work resulting from AI or not. Here you are 10 useful details for recognizing images generated with AI.

1. Incorrect fingers: artificial intelligence algorithms still struggle to generate fingers (as demonstrated by the photos in the post published on X that we included at the end of the article). As explained by the expert Amelia Winger-Bearskin, professor of artificial intelligence and the arts at the University of Florida, “AI doesn’t really understand what a hand is, at least not in the way it connects anatomically to a human body.” It is for this reason that images generated by artificial intelligence containing human characters, in some cases, are easily recognizable by the presence of hands with 6 or more fingers, inches absent, fingers that “fuse” togetherand so on.

   

2. Teeth anomalies: artificial intelligence sometimes also struggles to reproduce teeth, not making a distinction between incisors, canines, wisdom teeth, and so on: when the dental arch appears clearly regular and unnatural, it is likely that the image is artificial.
3. Deformed faces: in some cases the faces of the characters portrayed in the images may appear deformed or asymmetrical. This is a rather obvious detail, which immediately catches the eye and represents a clue to take into account.

   

4. Unnatural looking hair: another anatomical part that AI is not always able to reproduce is hair. Sometimes there is no coherence in the hair, which presents a varied and unnatural texture. In the most extreme cases, hair can even transform into an accessory (for example a scarf).

   

5. Unnatural skin texture: AI texture problems don’t just affect hair, but also skin. When this does not have those imperfections and details typical of human skin, and the character portrayed seems to be a sort of rendering, this is another detail to pay attention to.
6. Architecture errors: misaligned steps, strangely curved walls, sloping ceilings for no plausible reason, tables that stand upright despite only having one or two legs, are all typical errors in images generated with AI.

   

7. Errors in accessories and hand-held objects: the presence of deformed jewelery or strangely elongated or even suspended objects when held in the hand of the subjects represented in the images are details that sometimes characterize artificially generated images.
8. Artifacts: if the images present artefacts, such as missing limbs or the presence of three eyes (which can be noticed especially when many people are represented in the background), or in any case graphically inconsistent elements, it is likely that the image is the result of artificial intelligence.

   

9. Confusing text: the texts present in some elements represented in the artificial images (for example on road signs, shop windows, posters, etc.) may present artifacts that give them an unnatural appearance.

   

10. Other details to pay attention to: the algorithms that generate images could try to hide one or more errors among those listed in the previous points by blurring rather important areas of the images, such as the background. In other cases, users who generate such images may deliberately crop the photo so as not to show errors (such as those relating to fingers).

    

Software to recognize images created with AI

It must be admitted, however, that not everyone has a trained eye for spotting AI-generated images. Furthermore, the software that generates images is constantly being improved, which makes the task even more difficult. For this reason it could be useful to use specific online tools, which allow images that are suspected to have originated from artificial intelligence to be examined under their lens. Among the best on the market we remember AI or Not, FotoForensic and IlluminArty. Using them is quite simple: just upload the image you think may be artificial or paste the URL into their appropriate text field, start the examination process and wait for the response from their algorithm.