Eso's Guide to Identifying AI Generated Images

[esotericas 2674]

 

✩ IDENTIFYING AI GENERATED IMAGES ✩

 

•─────────────────────•✦•─────────────────────•

 

 

While I am personally very strongly opposed to the use of AI in image creation, I am aware that some others on LOTC are less opposed. I did not write this with the intention of trying to change anyone’s mind, just of trying to help people know what they’re looking at. I will not be preaching about my views on AI during this post, at least not intentionally. If some of my opinions leak through, please forgive me. 

 

Please don’t use this post as a place to argue over the ethics of AI. This is intended to be purely informational. 

 

TL;DR? 

ZOOM IN on images. Look at them for a second before you download/pin/save them. Most AI images look real at a glance. It’s only when you spend the time to stare at them that you can identify them as AI-generated. 

 

I

Terminology

 

Many AI images are prompted with broad, general labels. “Victorian,” “Medieval,” “Fantasy,” or “Sci Fi,” for example. These are more effective at producing a reflective image during generation, because the AI model has a vast amount more reference material to pull from than if you tell it, say, “1885” or “Early 15th century.” When posted, AI images are often captioned with the terms used to generate them, or with similar terms. Vague, general words like those seen above will therefore be much more likely to result in AI images.

For example, of the first ten or twelve images I received when Googling “Fantasy Dress Art,” I identified five as being AI generated. “Elven Dress Art” only produced two, and “14th Century Dress Art” produced none. If you don’t know what to call what you’re looking for, talk to a skinner, or send your references to me! I will always be happy to help people find what they’re looking for.

Many of the AI-generated images you’ll find in internet searches come from websites offering “Thousands of free PNGS” or something along those lines. Don’t trust anyone who offers you 50,000 images of ladies in dresses! Red flag! Check image captions on Google and Pinterest. Here are some examples of captions on AI images:

Now, not every image from these sites will be AI. Don’t use this as your only metric. But if you’re unsure about an image, and the captions look anything like those, it’s likely AI-generated.

 

II

Wonky Details

 

For reasons that are somewhat beyond me, AI is often very bad at repeating geometric shapes - I’d assume something to do with averaging lines. Whatever the reason, it can often generate “geometric” shapes that fall apart or blur together, like these, which are clearly supposed to be hexagons.

Admittedly, this is a rather extreme example, which I shared mostly to prove my point. Here are some subtler examples of AI “losing track” of what it’s supposed to be creating. 

 

Bodice lacing “wobbling” and merging into itself.

 (Also, take note of the low resolution of all of these images. While I’m not sure why, many AI-generated images can be very low-res.) 

 

Similarly, since AI isn’t consciously “designing” each image it makes, it often struggles to make the small, subtle details coherent. If you zoom in enough, all the pattern and detail becomes... seemingly random. It’s not clear what it is, or what it’s trying to represent. 

 

The left image is AI, the right is not. On the left - what is this embroidery? What are these patterns? We can’t tell. On the right, even though the image is low resolution, we can clearly see a pattern of stripes on top, and some sort of braid down below. Even without the context of the larger image, the detail is effectively communicated. 

Here are some more examples of meaningless detail in AI-generated images. There’s no way to know what these patterns are supposed to be.

 

Because AI can generate what would be hours of work in a split second, it can often generate more detail than a human being would bother to. It has no need to simplify. While a human being might have simplified or stylized the embroidery on this bodice into something easier to draw (and easier to understand) the AI doesn’t understand the need to do this, and produces an incredibly over-detailed dress. 

Another example - this is supposed to be a painting. No reasonable person would bother to paint this many wrinkles - and no reasonable person did. It’s AI.

 

III

Asymmetry

 

In a similar vein, AI can’t consciously monitor its own symmetry. While some real art will have intentional asymmetry, the asymmetry found in AI-generated images is often difficult to notice or “accidental” looking. For example, let’s take a look at laces-lady again. The left side of her bodice has a blob of lace that isn’t present on the other side. This adds nothing to the design - it doesn’t feel like a conscious choice... Which is because it isn’t one. These sorts of subtle asymmetries are everywhere in AI-generated images.

Why does this woman have an orange part on one sleeve, but not on the other?

Why are this woman’s sandals different? (Also, her toes...)

And this leads us to our last section. 

 

IV

Anatomy

 

This is the one you hear about, so I’m not going to belabor it a ton. As you can see with those freaky toes up there, AI really struggles with finer details of the human body. It’s gotten much better about hands, but feet, eyes, and ears are all often wonky or blurry. 

 

This queen is looking in two directions at once.

This one’s head isn’t centered over her feet.

And this little girl’s arms and legs are WAAY longer than they should be.

 

V

Byebye

 

Everything else there is to say about identifying AI-generation has been said a hundred times before - jewelry, hands, anime style, et cetera. These are the little tips and tricks on top of those standard tools that I use. If I can impart you with no other wisdom, I hope I can remind you of this:

ZOOM IN.

Zoom in. 

What looks reasonable at a glance looks uncanny when you stare at it for a second. It’s always worth spending the time to be sure you know what you’re looking at.

 

That’s all. Thank you for reading!

 

•─────────────────────•✦•─────────────────────

[squakhawk 14038]

1 hour ago, esotericas said:

 

 

its kinda like ai writing where u just get a feel for it. there might be "ai image detectors" but at least for writing the ai writing detectors are totally innacurate, you just sort of get a feel for the style. pinterest/artstation/google images are fucking destroyed by ai garbage so youre gonna get trained to recognize it one way or another 

 

searching on google lens will probably bring up the image or 10 similar images in slightly different poses (different generation results) so thats another easy way to see it

[TheBigBubbles 963]

I usually have a good feel for if something is AI generated, but even after looking at the picture of the child for about ten minutes, I still cant see it.

[teeylin 1688]

good soup

 

as an artist, it's so discouraging to see people type in a prompt into a website and generate massive amounts of "art" that would take me hours upon hours to make.. but at least i can draw hands. **** you and your hideous hands, ai

[Cheese 8882]

The day AI Habsburgs' itself out of existence is a day I will rejoice.

[sapphic_spidy 1089]

i once saw someone use an ai image for their character and on the side wrote in "by me"

that was not by you

the buttons are literally melting into each other

[Cheeseycereal 829]

9 hours ago, sapphic_spidy said:

i once saw someone use an ai image for their character and on the side wrote in "by me"

 

 

SPIDDY!!! IT WAS BY ME!!! HOW DARE YOU!!

[Gaja 1311]

it is really heartbreaking seeing ppl resorting to AI images for their creative writing when we have the vastest pool of beautiful, man-made art since the dawn of time at our fingertips. on social media like twitter you can find a LOT of artists who make really cool art that deserves more eyes on it

i think ai art is so prominent bcs ppl want images that directly correlate to what they are trying to portray but imo, why? in creative writing, you can use your words to effectively describe it and the image can be a really strong mood-setter instead. but that's just my POV

[sapphic_spidy 1089]

On 1/31/2025 at 1:13 AM, Gaja said:

i think ai art is so prominent bcs ppl want images that directly correlate to what they are trying to portray but imo, why? in creative writing, you can use your words to effectively describe it and the image can be a really strong mood-setter instead. but that's just my POV

people who use ai art also tend to be lazy and impatient when it comes to art

they want something now, they don't want to have to wait, they don't want to describe it beyond prompt words into a text box