Art generated by AI, or AI art, can look pretty normal at first glance, almost like an intermediate or expert artist drew it. But then you’ll see the details, or rather the bizarre details. Crisscrossing lines that imply fabric folding, but in reality mean nothing. Inconsistent shading. Eyes that either look insect-like or completely dead. These are just some of the tips for deciphering AI art.

For this article, I want to introduce to you some of the tells that AI art gives off. I’m a Digital Artist myself, and have 10 years of experience under my belt. While I would not consider myself an expert, I can still tell when an art piece is off.

Warning: Some of these tells aren’t just tells of AI art – they could also be signs of a beginning artist too. Mistakes are common, and are impossible to avoid for all artists.

While looking for a good source, I found Paige, or Duchess Celestia, an Artist and commentator on YouTube. As Duchess Celestia puts it in her video on spotting AI Art; “Human artists make technical mistakes. We make arms too long, we put eyes too close together.” Human artists are using their interpretation of what they are envisioning and/or seeing, and trying to draw that. Beginning artists may not have the skills or knowledge base to quite understand how fingers connect to palms, but it will still look like a hand. 4-5 fingers depending on style, the fingers bend at the knuckles, and the thumb is roughly where it’s supposed to be.

In comparison, AI makes “logical mistakes” and technical mistakes. According to Duchess Celestia, AI doesn’t think like we do, it doesn’t know what a hand is. To AI, 6 fingers and a thumb is just as good as 4 fingers and a thumb. The AI doesn’t know what a palm is, and certainly doesn’t know where it should be proportionally to a wrist. AI can only replicate what it’s been fed – not create something entirely new like a Human Artist.

Human artists have instinctive knowledge. AI can only generate based on the images it’s been fed.

Now, getting to the tells.

  • Hands are a good starting point. AI doesn’t know what a hand looks like. How many fingers does it have? Does it have more than one thumb? Are the hands logically connected to the body? Is it floating on someone else’s shoulder?
  • Look at the eyes. Are they looking at the same focal point? Are they roughly the same size and shape? Is the structure the same? (iris, pupil, sclera) etc.
  • Uniform and plastic textures. Artists eventually learn how to render different materials. Metal looks and shines differently than skin, hair shines differently than silk, and velvet absorbs light differently than cotton. Artists learn how to memorize and recreate the light absorption and reflection effect of different materials. AI doesn’t know how to do that. Skin is just as reflective as silk and metal as far as AI is concerned.
  • Lack of consistency. Consistency is difficult for a lot of artists to do, especially beginners, but AI does it differently. Nostrils will be uneven, earrings will look entirely different and muddled.  Dresses and shirts may have contrasting details that are not designed thoughtfully.
  • Blending of details. Oftentimes AI art pieces will have a blended airy effect. This may be a result of incorporating a lot of art pieces that reflect an airy, soft, and lighter art style, but that won’t explain earrings blending with hair. Hair blending into one massive blob. Or the hair blending into clothing or the background. That type of blending is not typically seen in an artist’s work, even beginners.
  • Confusing lighting. Where is the light source coming from? Does it make sense for the shading?
  • For artists of higher skill levels: think critically about the piece, and mentally reconstruct the drawing process. Does it make sense? Is it a logical process?
  • Strange composition and off perspective. Artists place elements in a picture to lead the eye around in a pleasing manner. An artist follows some rules of perspective. Items further back in the image are smaller, blurrier, or are more uniform in color depending on the art style. AI art doesn’t understand the idea of composition, it just places things randomly based on the images in its database. An AI doesn’t quite understand perspective. It often makes everything in the background a uniform blur or rendered to the same degree as the foreground. Either way – the AI image will be more confusing to look at than the art piece drawn by a human.

To further show the differences of an AI art piece, versus a human made art piece, I decided to redraw an AI art piece I found on Pinterest. The poster was Roeum Rithykun, who clearly states in their Pinterest username, and their “artist” tag at the bottom, that their images are AI art. I decided to redraw this piece they generated.

Pretty right? I went over and picked out the problems I saw rather quickly. I analyzed it for logical and technical issues that might appear in AI art.

The biggest issues, going from top to bottom-

  • The pig-tail on the left has the implication of sensible curls, but otherwise I can’t figure out what’s going on. The right side is a bigger mess of muddied details. Lines that don’t work to create a definitive form of any kind. This is a symptom of AI art, and AI’s can’t logically use lines to help create a 3d image in a 2d drawing. 
  • Her earrings are an offense to an artist’s sensibility. They are not identical, not even consistent with one another. They are messy and inconsistent. Beginning artists are unlikely to make a mistake like this. Especially when the earrings look so visually inconsistent with the rest of the picture.
  • Her necklace is the clearest sign of logical failings in the image. The necklace juts into the background. It’s malformed, has random notches like a tree branch, and seemingly cuts through the girl’s turtle neck to connect to the other side. This screams logical failings, as an artist would be highly unlikely to make even one of those mistakes.
  • The folds of her left facing sleeve are inconsistent, and imply a pillow-like texture. something not matched by the torso part of her shirt.
  • The shading of the entire shirt is not consistent with the shading done on the face. It’s almost comic book-like, with its primarily black, solid shading.

With all that in mind, not even counting the poster identifies it as AI, I can definitively say that this was made using AI.

So, with those notes in mind. I made this drawing. 

Many of the aspects I pointed out for the AI piece, I sought to correct or change to make a more cohesive drawing. Despite that, I still made mistakes. I made her hair texture more readable. Her scrunchies have a more defined shape. I fixed those earrings, but I realized my attempt to show that the flat star section and the spherical ball section could twist and turn independently of each other fell flat. I simplified the necklace, perhaps too much. It looks like a shoestring now rather than a metal necklace. I feel like I made the torso and sleeves different weights – making the torso heavier while the sleeves are more flowy and fluid almost. I made her expression more serious, and I think I messed up her anatomy on her lips and mouth, which is an issue I run into a lot in my own art. In hindsight I realize I could have made her hair texture more coily to better align it with the original AI art piece. 

These are not all the tells of AI art. And these details, when found alone, do not mean a piece was created with AI. An artist is capable of making similar mistakes, so always consider how many points of logical or technical failure are in a piece before deciding it was generated by AI. This shows that, despite my 10 years of experience as a digital artist, I can still make mistakes and so can other artists. That’s what makes us human and that’s what brings life to art.