With the growing trend of AI being used in almost every field of creative space nowadays, it’s unsurprising that it soon evolved into the idea of generating AI anime art. Of course, the entire concept of fabricating these works is already a set of controversies that are still ongoing. But if we detach ourselves from the negative ramifications of such technology, it is still a pretty nice tool. Definitely something that can be used for our nefarious purposes.
And so, we shall ask the question: can we use current LLM-based AI tools to generate kigurumi onesie art? Spoiler alert, the image above already confirms a massive yes. But, the implementation may not be as straightforward as you think.
DISCLAIMER: Before anything, I would like to explain first that these images are created using various free tools online with rather simplistic (non-paid) features and capabilities. This specific blog post is only meant to showcase the possibilities, and is not representative of a fully optimized method. Because more dedicated AI tools and better input prompts would definitely create much better images.
Polar Bear Kigurumi Onesie on a Sofa
This is probably the easiest to generate, with very simple prompts, and with the AI tool seemingly having a good understanding of what a kigurumi onesie actually is. The drawing itself isn’t that complicated, and the theme is also very straightforward.
There is, however, the obvious imperfection that is the leg anatomy of our subject. This is still quite common to simpler AI tools when not given enough prompts, or if the accuracy of the description is not established. Thankfully, you can still get a good idea of what is portrayed in the image, and you can perhaps turn half a blind eye to the aforementioned flaw.
Breakfast in Bird and Raccoon Kigurumi Onesies
This one also turned out to be somewhat good, with most prompts accurately portrayed in a way that the image does not instantly look wrong at a glance, or if you have no idea what prompts were used. Well… apart from the usual hand anomaly, the food items definitely look suspicious if inspected more closely. I mean, you can even tell what in the world are they exactly eating (or drinking?).
As specified by the title, the two kigurumi are supposed to be bird and raccoon-themed. The AI definitely got the beak, but still tried to make it mammal-like. The raccoon kigurumi onesie is completely unrecognizable, and we’re even sure what the AI tried to draw. Plus those two hoods are overlapping each other!
Licensed Character-ish Kigurumi Onesie… in a Car?
There are… lots of things very wrong in this picture. The original intention was to generate an image that is set inside the car, with the subject holding some sort of Starbucks-like product. Uhh, it did manage to capture these elements, but in a completely unintentional way.
First off, the hands. Like, the AI just completely threw anatomy out of the proverbial window and decided that it wants a cool pose while still having extra hands to hold the cup. And I thought the derpy curves they use to fabricate their idea of a hand was already weird enough.
Second, is their idea of a car interior. Apparently, this must be some sort of a specialty café where the window seats are themed with car doors. It must have mixed the Starbucks cup prompt with the car setting, creating this rather amusing visual anomaly.
Cool kigurumi onesie though, and its theme is the only element that survived intact.
The Missing Octopus Kigurumi and Non-existent Kitchen
So for this next art, I had the AI tool create a very generic kitchen setting, with the only unique element being the octopus kigurumi. Not only did it manage to completely miss the octopus theme entirely, but the kitchen doesn’t even seem to exist. Are you telling me that this is the AI tool’s idea of a kitchen? I tried other prompts that included the keyword “kitchen”, and they all turned including the kitchen as normal.
Also, while I love the unique design of the kigurumi onesie, this particular art style seems to have an obsession with cute mammalian motifs for kigurumi themes. I wonder what kind of objectional furry oddities would it re-interpret next?
Panda Kigurumi Onesie at a Beach Café
Ah yes, finally, an absolutely perfect illustration that encapsulates the ideas provided by my prompts. We see the entire profile of the onesie chosen for the image, and it even has a rather splashy color palette to boot!
…. Wait a minute. How in the world is that panda kigurumi colored blue? Is this some sort of sick reference to blue penguin kigurumi onesies since it is beach-themed? The AI did manage to understand the concept of a panda as a kigurumi theme, as it even has the signature ears and face. But it somehow blundered completely by giving it a very pastel-like shade of blue. Are we… are we looking at a theoretical licensed character kigurumi again?
Kigurumi Partygoers Dancing in a Park
Yes, a park.
They are supposed to be dancing in a park. A green, clean, park lit by the beautiful noon sky. Weird thing is, earlier results did get the park setting correctly. But the kigurumi onesies keep transforming into sports jerseys, as if the AI is insisting that the park only fits the setting of a traditional sports anime or manga series.
Instead, what we got is a studio-like area. All of them are still dancing though, so the most important theme is still preserved at least. But as you can already plainly see, the AI really got lazy at the end, as it just randomly doodled that person over the far side of the picture. (while still making sure that we can somehow understand that he/she is still wearing a kigurumi onesie)
Conclusion: More than Enough Potential
Actually, there were a good number of really good entries even when just using free AI tools out there. The only caveat... is that most of these AI understood kigurumi as hooded jackets, and not the onesies that we are more generally familiar with.
Despite these hurdles, we can see just be these early cursory examples that it is possible for LLM-based AI to understand the concept of kigurumi onesies. Perhaps we should go for a more realistic visual approach next time?
And maybe stop always using white or black hair to dabble on other hair colors.