A few days ago, I posted a big editorial all about the new DALL-E 3 and its freaky ability to create high quality ponies with even less effort than previous AI "art" methods. Since then I've been flooded with Discord messages, emails, comments, and DM's asking all sorts of questions about it, with many people giving me examples of work they had commissioned from unknown artists asking if it is AI.
This is kinda scary in its own right. The last thing I want to do is start a witch hunt against people just breaking into the MLP art scene. I do think it is super important to educate absolutely everyone on this technology as a whole though. As I mentioned previously, this model is the first ridiculously easy way to create just about anything. For fun yesterday, I lazily slapped a "luna pony praising the sun as Solaire from dark souls" into DALL-E, and while it didn't generate what I wanted at all, it did spit out this very pretty, albeit artistically messy picture of Luna:
The thing is, that was even less effort than a google search. I put 0 thought into it and just took the first thing it threw at me.
We already see loads of bad actors using stuff like this to make fake rage pieces and fool people into believing stuff that didn't happen. Educating yourselves on what is already here and what is possible in future versions of this and other AI iterations is the best defense. This is a MLP site though, so today I'll be focusing specifically on pony, so you can at least be somewhat aware of who you are commissioning or what YCH you are buying when trying out new artists.
Head on down below to learn what to look out for in cartoon equine specific DALL-E 3 Generations, and how to protect yourself.
 |
| I won't have a section on extra ponies, but getting DALL-E to draw groups is almost always going to result in obvious AI. |
It is extremely important that you don't go out there and start bothering every artist for every little error they do in their own work after reading this. Artists are human. It's what makes this medium so special. They are going to have wonky hooves, weird perspective issues, missing horns and wings, clothing that doesn't flow correctly, and loads of other minor detail problems that DALL-E 3 also has. Be educated on what the AI typically messes up, but also don't use this knowledge to attack anyone.
Anyway, on with the learnings!
The Obvious and Easiest One - Cutie Marks
>
DALL-E 3 can't cutiemark very well at all. Every once in a while it gets close, but it really isn't reliably capable of it. Even Luna's "inkblot with a moon" tends to fail horribly despite being one of the simplest and most variable.
This is also the absolute easiest thing to fix for even a non-artist. All it takes is being able to erase, blur, and slap a new one on and warp it to fit. A lot of artists do this anyway to save time, so if you see an obviously edited cutiemark that hasn't been properly melded into the drawing, don't immediately think AI, but use this simple "1st check" on an image to see if the butt symbol is accurate for a quick AI filter.
Horns Where Horns Shouldn't Be
Our buddy DALL-E loves to channel its inner M.A. Larson. Outside of mane cast and other easy to generate ponies (basically anypony with tons of fan art), it loves passing out horns. This is actually dodged somewhat for pegasus, but most OC's and backgrounders are going to have one half the time. And in a few cases, like in that Aztec Luna in the previous set, it gives alicorns two! This is also really common in bats for whatever reason.
Oddly enough, a lot of people that submit AI art to the EQD box as their own art don't even notice. They don't notice a lot of things actually. Use that against them.
Wing Errors. So many Wing Errors
DALL-E 3 actually generates pretty amazing wings typically... it just does too many. It also puts them in weird places. Lots of neck-rooted wings, wings appearing in the background, wings coming out of butts, wings from bellies, and when it does screw up, they are pretty disfigured like that Luna on the pillow up there. Bat wings tend to have feathers, and sometimes wings even absorb other aspects of the image like this one:
.jpg)
Hooves, the Bottoms in Particular
Ironically, this is something previous AI models like Purplesmart, Novel AI, and especially the newer furry stuff do phenomenally well. AI isn't always an upgrade in every area. DALL-E really struggles with the bottom of hooves. In fact, specifying it might end up getting you some pretty gnarly nightmare fuel. In many cases they come out as cups or weird cones where the hoof gets smaller as it ends rather than bigger around, like the examples up above. When sleeves are involved like that mad scientist Luna, they might not have hooves inside at all since they see the sleeve as a hoof.
You will see a lot of mangled legs and hooves if you browse people's DALL-E pony on Twitter or Deviant Art since non-artists don't seem to notice them at all. I've had a bunch of these come in with obvious AI legs bending in positions even newbie artists would never do.
That being said, hoof perspective can be rough, so don't always immediately assume AI on this one since even the best artists produce weird legs sometimes.
DALLE-3 produces very convincing eyes, but a lot of them have minor imperfections an artist might not usually make. Pupils in particular come in a wide variety of sizes and shapes. This one is basically impossible to spot at a glance though, so it's one of those final checklist items to go through when trying to figure out if a piece of art is a generation or not. I'm going to go through each of these one at a time because of that.
First off, the cyberpunk Luna with the magenta eyes. Eye color is something you can look for, but people draw ponies with different colored eyes all the time just for stylistic reasons, so I wouldn't use it. What really stands out in those eyes is the ridiculous fractal geometry in them. This is a VERY common thing with DALL-E, and rare for an artist to do. Usually you are going to have some kind of cohesive design between the pupil and iris. Not a bunch of colors spiked all over the place and random mirroring.
Our second nerdy NEET Luna has similar problems. it looks more like a badly cut gem than an eye, with the pupil and iris essentially merging.
The third one with Warrior Dark Souls Luna there is rocking a tadpole shape iris. Another common DALL-E 3 thing is weird shapes, and this is a perfect example of it.
And finally, probably the best eyes of the bunch is that cute wedding Luna. Pupils aren't fusing, ovals are ovals, and things are where they should be, but there is a ton of artifacting with the eyes overall. This one is tough though. Every artist in the fandom has a unique eye design, and without large glaring errors like the others in this set, you could argue that this is just a stylistic choice.
Honestly I wouldn't use that final one as an example of "obvious AI eyes" It's good to know how DALL-E can pull it off though even with a complex painterly design like that.
The pony might be flawless or near flawless, but the more complex an overall piece is, the more likely you will run into errors somewhere. The background is basically guaranteed to have some obvious derps if it has any kind of complexity at all, and when you give anyone the tools to create beautiful set pieces, they are going to do it and probably not notice the issues.
Luna's cobblestone street in that first one drives itself directly into the water, with floating railing. I guess you could argue there was a flood or something, but it's a stretch.
The second one could easily be argued as an artistic preference, so I wanted to include it as yet another warning not to bother artists for weird things in their drawings. You could say that some of those floating balls make no sense at all, but a abstract, trippy disco setting isn't exactly rare.
The third one is pretty much just object derps. A piece of the fence is floating in the sky for some reason, and while most artists are going to do very simple looking background elements, the pony figures on the book shelf are very AI-like.
And finally, witchy Luna's background has all sorts of oddities. A lot of the candies make no sense at all, fuse together, or have broken text on the wrappers. Text isn't too common so I'm not going to have a section for it here, but it's almost always misspelled and broken unless it's something really simple or a known logo.
Unfortunately, DALL-E 3 can definitely mask a lot of this just throwing out unique styles and basic images. If it doesn't have a background, cutiemark, hooves, complex eye shading, or anything else in this post the only thing you can really do is zoom way in and see how the lines work.
Android Luna's seams are a little wonky on the arm and belly, but telling an artist a few of their lines suck is a good way to look like an asshole.
The mafia batpony's lower teeth are a little strange, but again, it's so minor it's kind of irrelevant.
Mafia Luna is a more complex image, but the stars just aligned there. You could say her left hoof is disappearing, but you could also argue the sleeve is just rolled forward from the post. You could question why she's wet despite being indoors, but she might have just gotten out of the rain and is still dripping while she hops up on the bar for her drink.
And when it comes to OC's and Luna, cutiemarks are all over the place. I've seen hundreds of variations of her butt tattoo over the years. I wouldn't bat an eye at someone simplifying it like that.
That's the big question isn't it? Can humanity even handle being able to produce anything digitally instantly? I made these posts partly because I'm tired of people getting hacked and scammed on Discord from "artists" direct messaging them out of the blue with AI art asking for commissions, but also because I genuinely think it's important that everyone be aware of what Large Language Models are capable of.
A year ago DALL-E was a joke. Now it's not. In the last 365 days we've already seen fake pope fashion headlines, VA's complaining about Skyrim modders using their voices for AI NPC's, and politicians freaking out about deepfakes. Everyone needs to be educated on this topic so we can better prepare for the next generation of this tech. The DALL-E 5 that can perfectly produce a pony in any possible situation with no errors in any style.
Take this knowledge and apply it outside of pony. If an enraging picture on social media pops up, question it before you react. Be aware that we are already there, and it's only going to harder to tell what is and isn't built from a prompt thrown into a LLM.
