Ever since AI came out, I’ve been pretty suspicious of it. While quite a few of my friends assure me that it’s fine and can be a wonderful tool, I still stay wary of it. Can it be used for good things? Yeah. But can it also be used for evil and be really creepy too? Also, yes. I mean, according to one of my friends, the more you use Chat GPT, the more the system adapts to respond in a way that corresponds to your personality.
My suspicion of AI still lasts to this day, but over the weekend, I decided to play around with a design app that I got on my new computer that happened to have AI. Basically, how it works is you just type in a prompt that’s as vague or specific as you want, hit generate and it generates the image that it thinks you want. Intrigued by this, I proceeded to spend pretty much an entire day playing around with AI art, trying to figure out how it worked and seeing if it could generate anything that I could use, so here’s my review of AI art.
#1: Human Faces
For some reason, when you ask AI for art that depicts a human or human-like figure, most of the time, what you’ll get is something that looks somewhat realistic, like anime, or a weird mix of both. While I did observe that there were some variations in style, anime or super-airbrushed realism were the most common, which means that the faces would generally look pretty similar. While this could’ve maybe been fixed by changing the prompts more, a lot of the pictures that I got were really similar to previous ones and had the same issues. For example, for both guys and girls the lips and cheekbones were way too prominent to the point where people looked like they were going on a mewing streak and if you added in something like a specific eye color, the eyes became piercing to the point of it being scary. They just didn’t look natural at all. While these faces are definitely better than I could draw (I need a template to draw most things that are geometric shapes, honestly), there’s just something about them that completely give it away that it was done using AI. For example: these two pictures…
Not only are there some obvious problems in clothing design (such as how the first boy’s axe looks kinda awful and the armor could be better), but the faces look quite similar to each other; not exactly alike, but similar. Throw in the super airbrushed look, and you have an image that was absolutely made with AI.
(I also just want to throw in here that my original prompt for the first boy included “close-cropped brown hair” in the description. Nothing about his hair in the photo looks even close to what I specified except for the color. The second boy was also supposed to have dragon wings, not an actual dragon. On the bright side, at least I can use these images here without worry about people stealing my character designs as AI depicts them.)
#2: Other Body Parts
This has been a common complaint with AI ever since it came out, but if you try to generate an image of a human, the main part of the human may look fine, but things like feet, hands, wings (if you’re doing a fantasy human-fae thing), etc. can come out looking terrifying. While I was having fun generating characters that I’ve created, when I was trying to give a character wings from a specific creature – like a bird or a dragon – what would often happen was the animal’s head would appear on the wings or there would be no wings at all and it would just have the animal sitting on the person’s shoulder or something. Another example of AI being bad at generating limbs is when I typed in the prompt, “Jesus flipping over tables” and this was the result:
First of all, that’s hardly a flip and secondly, what is going on with his hands? On the left hand, he has ET fingers, and on the right hand it kinda looks like he might have an extra finger. I’ve also seen other AI generators have problems generating feet to the point where they look like they’re better suited for a chameleon, so that’s definitely a glitch that will need to be worked out.
#3: Didn’t Know What a Tanuki Is
This is obviously my biggest complaint with this AI software. For several hours, every time I told it to generate a picture of a tanuki, the tanuki almost always came out looking like a red panda or a raccoon to the point where I had to teach the AI what a tanuki looked like to get it to generate something even close to what I was looking for. Thankfully, after a few more attempts, I managed to get it to generate something that looks good, which leads me to this part of the post…
#4: Which Tanuki Should be the New Mascot for the Tanuki Corner
Here are the best images of tanuki that I got AI to generate. Please vote on which one you would like to see be the new mascot for the Tanuki Corner in the comments below.
#1: Somewhat Modern Tanuki
#2: Tanuki with Cherry Blossoms
#3: Chibi Tanuki
#4: Realistic Tanuki
Until next time,
M.J.
The Tanuki Corner 
Leave a comment