In a modern world where artificial intelligence has already generated as many images as photographers have taken in 150 years, the line between real and AI-created content has become increasingly blurred. However, for those who regularly work with AI models, this is not a big concern. As you come across lots of AI-crafted images on a daily basis, you may naturally develop the ability to recognize them effortlessly. Within our Everypixel team, we also have experts who work a lot with AI-generated images, and they’ve learned how to tell them apart from human-made ones. In this article, we share their insights and tips, guiding you on how to spot AI-generated images.
AI-generated content can often be detected at first glance as certain visual defects serve as clear signs of AI involvement. Our colleague and photographer Aleksandr Seledchik, a frequent user of Midjourney, has developed an almost impeccable skill for distinguishing between AI and human-crafted images. Here, we unveil some less obvious details which deserve your attention:
AI algorithms sometimes generate objects that stand out due to their unusual size or noticeable asymmetry. Look for buildings with distorted proportions, body parts with distinctive asymmetry, or even crooked frame structures.
In this image, for instance, a noticeable discrepancy in shoulder size creates an asymmetrical and disproportionate appearance.
When different elements are combined within an image, AI-generated content may exhibit composition fusion, resulting in visible seams, abrupt transitions, or merging of elements between objects.
For example, here, at the bottom right, you can see how the bird’s beak has merged with the petal.
Hairstyle have a lot of variability and details, making them one of the most difficult things for an AI to capture. Keep an eye out for unrealistic hair patterns, disconnected strands of hair, or a peculiar glow around it.
In the image below, you can notice a girl on the right with her hair growing from her hat.
AI-generated images might sometimes present unnatural hand positions, where fingers appear distorted or in physically impossible poses. Carefully examine hands and fingers in the image for such anomalies.
If you look at the image and precisely at the fingers on the left hand, you will spot this defect.
While the problem may soon be solved with the development of new AI models specially trained for text, it still exists and can help in distinguishing AI-generated images. Pay attention to instances of unreadable, nonsensical, or overlapping text within the image. AI also struggles to maintain consistent fonts, styles, sizes, and spacing throughout an image. In some cases, the text may lack coherence or clarity. Such artifacts are especially evident on small objects and in the background.
As an example, take a look at a photo of a police officer. The text on his badge is garbled, and it is just impossible to read.
AI creativity knows no bounds, and this is precisely what can help to distinguish AI-generated images. In the realm of factual accuracy, AI algorithms often craft settings that are difficult to imagine in the real world, incorporating elements that do not even exist and depicting super creative concepts. Here are some key points:
AI models can sometimes generate elements that look surreal or out of place. Such fictional artifacts might include familiar animals in unusual colors, non-existent devices or places, or even super extraordinary things like a cheese-shaped bus or a plane with a transparent floor. Keep an eye out for any suspicious items and check their existence in reality.
In this picture, the color of the birds’ beaks may not immediately stand out, but it is a clear sign that the image was created by AI. In reality, such birds with white beaks do not exist.
This defect is almost unnoticeable in portrait photographs but becomes remarkably evident in images where a person looks away. This sense of emptiness often evokes weird feelings, intensifying the “Uncanny Valley” effect within the photo.
Take a closer look at the woman’s eyes in the photo to notice its artificiality.
AI-generated content sometimes combines elements that would rarely or unlikely coexist in real life. This might include merging different clothing styles, architectural designs, or even details that are out of place in terms of historical context and culture.
One of the most striking examples is “Balenciaga Pope”, the image that was viral in March’23. At first glance, the picture looks real. However, it’s highly unlikely to see the Pope wearing such a jacket in real life. And if you take a closer look, you’ll notice some odd details, like the blurry engraving on the cross and the unusual choice to wear it over a down jacket. Uncovering these anomalies can be quite challenging, especially when a famous person is depicted in a photo. It is like our minds tend to automatically associate familiar faces with authentic photos, even when that may not be the case in reality.
These artifacts may often slip past the casual observer’s gaze, yet photographers never fail to spot them. Here we take a fresh perspective on AI-generated photos, exploring them from a less obvious angle:
The presence of exceptionally artistic and intricately detailed lighting can serve as a sign of a picture’s artificial origin. Images with diverse shades and tones often pose a challenge to recreate them in reality.
Here we compared the real photo with the one generated by AI. The first one, taken by a photographer, has consistent and natural lighting. However, when we employed Midjourney to recreate a similar scene, the resulting image gained complexity in shadows and cinematic effect.
Overly Smooth Textures
AI’s limitations extend to how it portrays textures, especially in skin, hair, and clothing fabrics. Pay attention to surfaces that appear unnaturally smooth, plastic-looking, and have a glossy sheen. These anomalies can be clear indicators of AI involvement.
In this image, the hair and skin appear remarkably flawless and lack natural imperfections. Such exceptional smoothness is a sign of the image’s artificial nature.
While AI continues to advance and produce increasingly convincing imagery, it’s essential to navigate these digital landscapes with a critical eye. In case of doubt, do not just focus on one detail — look at the whole picture. In most cases, it’s a combination of artifacts that reveals the image’s unreality. Keep practicing, delve into the details of a picture, and, above all, trust your common sense.