Hello! There's been an increasing amount of AI art on TBT whether intentional or not, so I decided it might be handy to make a comprehensive guide (with visuals!) on how to spot if a piece “artwork” was AI generated or traced over AI*. As an artist myself I do not consider these images art, so in this guide I will be referring to AI art as simply “images” as often as possible.
It should be noted that if an image exhibits one or few of the qualities listed here, it does not necessarily mean that it is inherently AI generated. Artists have different skill levels and styles. Many mistakes made by AI are made by beginner artists as well, so please use this guide as a tool to hone your own judgement.
*“Traced over AI”, refers to when someone generates a piece of AI art and then traces over it. This still falls under the AI art category to many artists because the base image still had to be generated to be traced over.
DISCLAIMER: I did not generate any of the images in this thread - I found them simply by Googling. I am also not a moderator and cannot force you to not like or use AI art, I just suspect that many people would not use an image if they knew it was AI.
Image Sourcing
Start with reverse searching every piece of art you save, and if you don’t have a source, don’t repost that image! This should be common practice whether the piece is AI or not - credit your artists.
You should also consider the website that you found the image on - Google now promotes a lot of websites that are solely used for AI/AI images, so be on the lookout. If you found it on social media, look at the user who posted it - some will indicate that they use AI in their bio - as well as any hashtags. Add “-ai” to the end of your searches to filter your search a bit. Example: If you wanted an image of Tom Nook, you would search “Tom Nook -ai” without quotes.
Sites that reverse image search: Google | Yandex | Tineye
Timelapses
I wanted to make a note of timelapses here to actually say to pay them no mind. In the very early stages of AI, a timelapse would tell a human artist apart from AI. But AI has since learned how to fake timelapses, so they should never be seen as definitive proof. Some programs also allow you to remove layers from timelapses, so people tracing over AI generated art can deceive their audiences that way. Please do not view timelapses as proof of anything.
Human VS AI Logic
The first thing to consider when looking at art is logic and intention - and that AI does not have any.
The reason so many people fall for AI is because they are only viewing the image as a whole. If you take a quick glance at an image made with AI, it might look incredible. But when you look at the small details, that’s when it becomes obvious that something is off.
When art is made by a human, every single detail is made intentionally. Even if we accidentally spill some extra paint on our canvas, we find a way to intentionally work with our mistake. AI cannot replicate this. When looking at art made by a person, you can see the thought process, the “bones” of the image so to speak. When drawing an arm, a human will think “Shoulder, bicep, tricep, elbow, flexors, wrist…” whereas AI will simply see the prompt “Arm” and spit out what it believes an overall arm looks like. The end result is often something illogical because AI does not have the capability to know what an arm actually is.
If most of an image is beautiful, then the artist is clearly skilled. So the easiest way to tell when an image is AI is to look for mistakes that a skilled artist simply wouldn’t make.
Below is a list of typical qualities you will find in AI generated images that defy human logic.
Soft, Dreamy Appearance
Have you ever noticed so many AI images kinda look like That, and you can’t really explain what That is? AI images tend to look so dreamlike and surreal for a couple of reasons, one being AI doesn’t understand texture. According to AI, skin, clothing, hair, metal, and wood are all the same. This is why surfaces made by AI look so excessively smooth and artificial.
AI also has no concept of background-middleground-foreground, so the image looks flat and like everything blends into each other, without any sense of dimension.
Evidently this person wanted a more realistic-looking Pikachu, but since AI doesn't understand texture, the fur lines go in every direction and overlap each other in some areas. His cheeks are also distinctly different textures, and the background can't decide if it's lightning, water, fire, or glass. Also, someone please give Pikachu his fingers and toes back.
Hands / Fingers / Limbs
This is the easy one. If a person has too many or too little fingers, the image was likely AI generated. Look at how the hand connects to the wrist, and how the fingers connect to the palm as well - AI doesn’t understand how body parts are connected to each other. Sometimes there will even be an extra/floating hand in an image. Look out for paws and feet in this sense as well.
Two left arms and an extra finger on the right hand, behold your typical AI mutant.
Aside from the cat's very obvious shark-fin ear, the paws are also inconsistent - the right barely has toes, but you can tell there's supposed to be four. Meanwhile the left paw has two very distinct toes, knuckle and all. The back-left paw also seems to be jutting out from the pelvis, with no actual leg in between.
*This rule of course does not necessarily apply if a character regularly has an unusual amount of hands/arms, so do your research.
Eyes
Eyes are another big tell. An AI generated pair of eyes will often have two eyes that don’t look like they belong together - The irises, pupils, or the eyes themselves are often different shapes/sizes. Pupils are sometimes squiggly, and some eyes may even have several pupils.
This Miku almost had me convinced - until it didn't. Her hair accessories and floating right ear may be the most obvious part, but let's look at her eyes. Her left iris warps into her upper lash line - which has no eyelashes on it, unlike her right eye. Her pupils are different shapes, and the right one starts to warp at the top. The bottom line of her right eye warps upward, where her left is more naturally curved. Notice the fake watermark as well.
Anatomy Continued
Going back to the point about arms and logic, anatomy is often wonky in AI images since AI doesn’t take the preparation steps that we do before rendering a person. When humans draw other humans, we first block out shapes, draw a stick-figure base, or draw the naked “barbie/ken doll” base before adding features and clothing. AI does not do this - it generates the entire image at once. When generating someone wearing a dress, it does not take into account the actual body under the dress, and when generating a torso, it does not take into account the room needed for a ribcage or internal organs.
Look at teeth as well - when you ask AI to make someone open-mouth smile, there are often three or four rows of teeth because AI does not know how many teeth someone is supposed to have, just that there should be teeth.
Clothing
Many times, it will be difficult to tell where clothes start or end. Earrings will not be identical to one another. Jewellery and other tiny details will likely be muddied together with their surroundings. Fold and flow lines will be inconsistent, and clothing will often look like different sections are made of different materials. Even with more complex clothing such as ornaments and armour, everything should still look like there was a thought process behind each section, and that you could create a 3D model of it.
I've seen many bad AI Links in my days, but this one takes the cake. The pattern on his tunic is inconsistent, and the tunic itself blends into literally every surface and object around it (the holster buckle, the scabbard, the undershirt, the background, his hair, his skin). His eyes are different sizes, and his right eye is rounder. The irises are blobs rather than circles. His fingers all blend together. Also, he's stabbing himself in the head?
Inconsistency & Asymmetry
AI has trouble with patterns and symmetrical lines. If there’s a tiled floor, a window, or patterned wallpaper, look at the lines and check if the pattern is actually continuous. Also look to see if objects outside of the window actually continue properly across window panes, they will often be displaced or cut off. This doesn’t only pertain to actual printed patterns either - chairs may have 4 different legs for example.
At a quick glance, this "Shadow Queen Peach" may look beautiful. But zoom in and look at the small details: her earrings are not symmetrical, her hair blends into the background in more areas than one, the floor tiles and skylight panes have random warping, the candles in the chandeliers are inconsistent and don't appear to be held up by anything, her left hand has a finger coming out of her palm, and right hand's ring and pinky finger appear to be webbed together.
Objects Blending Into Each Other
AI doesn’t understand tangents (when the lines of two different objects touch each other in a way that looks like they share a surface) so it will just blend the objects together. Hair turns into a necklace, limbs blend into each other, accessories blend into the skin, etc.
Lighting
AI generated images often have dramatic, high-contrast lighting and shading. That means the shadows are very dark, and the lights are very bright.
The other thing with lighting and AI is that it’s never consistent. This is a common flaw of many artists as well, so be cautious. There should be a clear light source or direction that light is travelling. AI doesn’t know that, nor does it know what a light source even is, so there will often be random spots of light from an unknown source.
This little guy is so cute! Unfortunately he is also a hot mess. The typical contrasted lighting gave him away immediately, and there are so many light directions in this image I lost track when trying to count. Other details to note: The rings on the underside of his cap don't continue all the way around and instead start going off track in the front. The leftmost red mushroom cap is floating, the stem on the mushroom to the immediate right blends into the grass, and the top-right mushroom's stem is sprouting from the cap below it rather than the ground. The rightmost mushrooms also have those white dots for an unknown reason. The two red leaves at the bottom are attached to each other.
Lines
Inconsistent linework, lines blending into the colouring, and lines that don’t connect to anything. AI can’t think of how to use lines to give form to a 3D object in a 2D style like we can. You may have noticed in the Miku and Link images that the lineart sometimes fades into nothing, or the lines of one object blend with the lines of another object.
Lettering
Words and letters in AI images have gotten better over time, but they’re still more often than not a jumbled and/or misspelled mess, like in the Miku image above and the "Solid Snake" you're about to see below. Sometimes they’re just nondescript squiggles. Be on the lookout for fake watermarks as well.
Eye-Level, Head-on Perspective
This is the most “not inherently AI” rule, but for the most part AI is terrible at perspective and will more often than not display its subject at eye-level. So, when combined with the other aforementioned qualities, you likely have something that was made by AI.
Wow, this AI Solid Snake looks so handsome! He also looks like absolute garbage! His armour and his facial hair are asymmetrical (did the razor run out of batteries?), his hair suddenly turns cartoony, the fabric and armour textures attach to each other randomly and with no logic, the shoulder strap buckles attach to nothing, the word on his headband is warped and not an actual word, and you can see his eye through the eyepatch.
And the funniest little tidbit about this image in particular is that in the post I grabbed this from, the user specifically asked the AI to generate Solid Snake. The only physical element that distinguishes Solid Snake from his father, Big Boss, is that Snake doesn't have an eyepatch. So AI didn't even get the character right and instead created a weird, uncanny mishmash of two of them.
Turns out, AI "art" isn't as hard to spot as its enthusiasts claim; you just have to know what to look for.
That is it for the actual guide! I am happy to clear up anything in this very wordy post, and I welcome suggestions and questions.
Since finding non-AI images for reference images has become increasingly difficult as others have mentioned, I've added some resources!
I will add more as I find more, and if anyone else has any more tools/resources feel free to post!
It should be noted that if an image exhibits one or few of the qualities listed here, it does not necessarily mean that it is inherently AI generated. Artists have different skill levels and styles. Many mistakes made by AI are made by beginner artists as well, so please use this guide as a tool to hone your own judgement.
*“Traced over AI”, refers to when someone generates a piece of AI art and then traces over it. This still falls under the AI art category to many artists because the base image still had to be generated to be traced over.
DISCLAIMER: I did not generate any of the images in this thread - I found them simply by Googling. I am also not a moderator and cannot force you to not like or use AI art, I just suspect that many people would not use an image if they knew it was AI.
The Guide
Image Sourcing
Start with reverse searching every piece of art you save, and if you don’t have a source, don’t repost that image! This should be common practice whether the piece is AI or not - credit your artists.
You should also consider the website that you found the image on - Google now promotes a lot of websites that are solely used for AI/AI images, so be on the lookout. If you found it on social media, look at the user who posted it - some will indicate that they use AI in their bio - as well as any hashtags. Add “-ai” to the end of your searches to filter your search a bit. Example: If you wanted an image of Tom Nook, you would search “Tom Nook -ai” without quotes.
Sites that reverse image search: Google | Yandex | Tineye
Timelapses
I wanted to make a note of timelapses here to actually say to pay them no mind. In the very early stages of AI, a timelapse would tell a human artist apart from AI. But AI has since learned how to fake timelapses, so they should never be seen as definitive proof. Some programs also allow you to remove layers from timelapses, so people tracing over AI generated art can deceive their audiences that way. Please do not view timelapses as proof of anything.
Human VS AI Logic
The first thing to consider when looking at art is logic and intention - and that AI does not have any.
The reason so many people fall for AI is because they are only viewing the image as a whole. If you take a quick glance at an image made with AI, it might look incredible. But when you look at the small details, that’s when it becomes obvious that something is off.
When art is made by a human, every single detail is made intentionally. Even if we accidentally spill some extra paint on our canvas, we find a way to intentionally work with our mistake. AI cannot replicate this. When looking at art made by a person, you can see the thought process, the “bones” of the image so to speak. When drawing an arm, a human will think “Shoulder, bicep, tricep, elbow, flexors, wrist…” whereas AI will simply see the prompt “Arm” and spit out what it believes an overall arm looks like. The end result is often something illogical because AI does not have the capability to know what an arm actually is.
If most of an image is beautiful, then the artist is clearly skilled. So the easiest way to tell when an image is AI is to look for mistakes that a skilled artist simply wouldn’t make.
Below is a list of typical qualities you will find in AI generated images that defy human logic.
Soft, Dreamy Appearance
Have you ever noticed so many AI images kinda look like That, and you can’t really explain what That is? AI images tend to look so dreamlike and surreal for a couple of reasons, one being AI doesn’t understand texture. According to AI, skin, clothing, hair, metal, and wood are all the same. This is why surfaces made by AI look so excessively smooth and artificial.
AI also has no concept of background-middleground-foreground, so the image looks flat and like everything blends into each other, without any sense of dimension.
Evidently this person wanted a more realistic-looking Pikachu, but since AI doesn't understand texture, the fur lines go in every direction and overlap each other in some areas. His cheeks are also distinctly different textures, and the background can't decide if it's lightning, water, fire, or glass. Also, someone please give Pikachu his fingers and toes back.
Hands / Fingers / Limbs
This is the easy one. If a person has too many or too little fingers, the image was likely AI generated. Look at how the hand connects to the wrist, and how the fingers connect to the palm as well - AI doesn’t understand how body parts are connected to each other. Sometimes there will even be an extra/floating hand in an image. Look out for paws and feet in this sense as well.
Two left arms and an extra finger on the right hand, behold your typical AI mutant.
Aside from the cat's very obvious shark-fin ear, the paws are also inconsistent - the right barely has toes, but you can tell there's supposed to be four. Meanwhile the left paw has two very distinct toes, knuckle and all. The back-left paw also seems to be jutting out from the pelvis, with no actual leg in between.
*This rule of course does not necessarily apply if a character regularly has an unusual amount of hands/arms, so do your research.
Eyes
Eyes are another big tell. An AI generated pair of eyes will often have two eyes that don’t look like they belong together - The irises, pupils, or the eyes themselves are often different shapes/sizes. Pupils are sometimes squiggly, and some eyes may even have several pupils.
This Miku almost had me convinced - until it didn't. Her hair accessories and floating right ear may be the most obvious part, but let's look at her eyes. Her left iris warps into her upper lash line - which has no eyelashes on it, unlike her right eye. Her pupils are different shapes, and the right one starts to warp at the top. The bottom line of her right eye warps upward, where her left is more naturally curved. Notice the fake watermark as well.
Anatomy Continued
Going back to the point about arms and logic, anatomy is often wonky in AI images since AI doesn’t take the preparation steps that we do before rendering a person. When humans draw other humans, we first block out shapes, draw a stick-figure base, or draw the naked “barbie/ken doll” base before adding features and clothing. AI does not do this - it generates the entire image at once. When generating someone wearing a dress, it does not take into account the actual body under the dress, and when generating a torso, it does not take into account the room needed for a ribcage or internal organs.
Look at teeth as well - when you ask AI to make someone open-mouth smile, there are often three or four rows of teeth because AI does not know how many teeth someone is supposed to have, just that there should be teeth.
Clothing
Many times, it will be difficult to tell where clothes start or end. Earrings will not be identical to one another. Jewellery and other tiny details will likely be muddied together with their surroundings. Fold and flow lines will be inconsistent, and clothing will often look like different sections are made of different materials. Even with more complex clothing such as ornaments and armour, everything should still look like there was a thought process behind each section, and that you could create a 3D model of it.
I've seen many bad AI Links in my days, but this one takes the cake. The pattern on his tunic is inconsistent, and the tunic itself blends into literally every surface and object around it (the holster buckle, the scabbard, the undershirt, the background, his hair, his skin). His eyes are different sizes, and his right eye is rounder. The irises are blobs rather than circles. His fingers all blend together. Also, he's stabbing himself in the head?
Inconsistency & Asymmetry
AI has trouble with patterns and symmetrical lines. If there’s a tiled floor, a window, or patterned wallpaper, look at the lines and check if the pattern is actually continuous. Also look to see if objects outside of the window actually continue properly across window panes, they will often be displaced or cut off. This doesn’t only pertain to actual printed patterns either - chairs may have 4 different legs for example.
Objects Blending Into Each Other
AI doesn’t understand tangents (when the lines of two different objects touch each other in a way that looks like they share a surface) so it will just blend the objects together. Hair turns into a necklace, limbs blend into each other, accessories blend into the skin, etc.
Lighting
AI generated images often have dramatic, high-contrast lighting and shading. That means the shadows are very dark, and the lights are very bright.
The other thing with lighting and AI is that it’s never consistent. This is a common flaw of many artists as well, so be cautious. There should be a clear light source or direction that light is travelling. AI doesn’t know that, nor does it know what a light source even is, so there will often be random spots of light from an unknown source.
This little guy is so cute! Unfortunately he is also a hot mess. The typical contrasted lighting gave him away immediately, and there are so many light directions in this image I lost track when trying to count. Other details to note: The rings on the underside of his cap don't continue all the way around and instead start going off track in the front. The leftmost red mushroom cap is floating, the stem on the mushroom to the immediate right blends into the grass, and the top-right mushroom's stem is sprouting from the cap below it rather than the ground. The rightmost mushrooms also have those white dots for an unknown reason. The two red leaves at the bottom are attached to each other.
Lines
Inconsistent linework, lines blending into the colouring, and lines that don’t connect to anything. AI can’t think of how to use lines to give form to a 3D object in a 2D style like we can. You may have noticed in the Miku and Link images that the lineart sometimes fades into nothing, or the lines of one object blend with the lines of another object.
Lettering
Words and letters in AI images have gotten better over time, but they’re still more often than not a jumbled and/or misspelled mess, like in the Miku image above and the "Solid Snake" you're about to see below. Sometimes they’re just nondescript squiggles. Be on the lookout for fake watermarks as well.
Eye-Level, Head-on Perspective
This is the most “not inherently AI” rule, but for the most part AI is terrible at perspective and will more often than not display its subject at eye-level. So, when combined with the other aforementioned qualities, you likely have something that was made by AI.
Wow, this AI Solid Snake looks so handsome! He also looks like absolute garbage! His armour and his facial hair are asymmetrical (did the razor run out of batteries?), his hair suddenly turns cartoony, the fabric and armour textures attach to each other randomly and with no logic, the shoulder strap buckles attach to nothing, the word on his headband is warped and not an actual word, and you can see his eye through the eyepatch.
And the funniest little tidbit about this image in particular is that in the post I grabbed this from, the user specifically asked the AI to generate Solid Snake. The only physical element that distinguishes Solid Snake from his father, Big Boss, is that Snake doesn't have an eyepatch. So AI didn't even get the character right and instead created a weird, uncanny mishmash of two of them.
Turns out, AI "art" isn't as hard to spot as its enthusiasts claim; you just have to know what to look for.
That is it for the actual guide! I am happy to clear up anything in this very wordy post, and I welcome suggestions and questions.
Non-AI Photo References
Since finding non-AI images for reference images has become increasingly difficult as others have mentioned, I've added some resources!
- Animal photo search tool with interactive, posable animal skulls *
- Animal photo references by Taxa
- Chickensmoothie's Reference Gallery (mostly animals, some humans and plants)
- Copyright-free image gallery masterpost by John Coulhart
- iNaturalist (animals, insects, plants/fungus)
- Wikimedia Commons
I will add more as I find more, and if anyone else has any more tools/resources feel free to post!
Last edited:
