Back Again, Just to Leave…

Wow, I’m just doing a bang up job blogging aren’t I?

Okay, let’s just say I’m trying to get better. How will I get better? Well, I’m turning art blogging into a hobby now I guess while taking a break from gaming which I am low key addicted to.

So news, my family is moving, leaving the country in fact, it wasn’t exactly safe here to begin with and now it is much worse. More on that in a few months. I’ll be blogging from outside the U.S. in August.

Oh, I also found a few people playing with A.I. and trying to get it to make a difficult “Where’s Waldo“ picture, they mean wimmelbilds but they don’t know what they are actually called. Wimmelbilds are somewhat niche. Either way I will let you look at the better results (let’s just say I’m not really all that concerned):

These were created by generative A.I. and not a person and not prompted by individuals who have an expertise in wimmelbilds (not their fault, just a fact). Beyond that, I don’t think anyone here was attempting to fix the output; its just the “straight out of the box“ generation.

Since I make wimmelbilds both digitally and traditionally, collect wimmelbilder books, study and use the works of wimmelbilder artists as reference I wanted to see what A.I. can do in this space.

My hot take around the criticism around A.I. art is this: it’s the human creator which allows us to approach art in a subjective way, without a human creator, it is just an algorithmically created image and can be criticized and judged in an objective way. The algorithm does not “improve“ on skills it gets better data to work with, gets more powerful etc.

My critique: These frankly suck. Which sounds harsh but I can say that with confidence and objectivity because, not much “thought“ was put into these prompts and as the human writing the prompt you end up being playing the role of the under-qualified art director with a poor understanding of the art you are asking for. Unable to define what you actually want and helpless to fix the mistakes or worse blind to the actual problems because you don’t understand what makes “good“ art. The more you know about art, the more discerning you’d become and you’d come to a similar conclusion.

So even output the average “A.I. Art Enjoyer” may think is “is very nice!“, is still unusable. There are no stories or little interactions between the figures in these images. The figures feel stiff, lifeless, and lack emotion just like the creator, ChatGPT, the composition and color choice is very uninspired and each image feels like just an excuse to shove as many figures into the composition.

Thanks A.I., I hate it. Instead of looking at cool or funny interactions, I’m playing “find all the mistakes”. So this one fails too. If you would almost spend as much time fixing the mistakes as you would just making this art with a different process, it loses any time advantage you thought you had, plus how many iterations would it even take to get somewhere here? This somewhat cleaner output is still too far away from being good to be worth the power used to generate it.

I have no words. Not really, listen, when you look up close to a wimmelbild and things are even harder to discern, you’ve failed at making a wimmelbild. Gemini fails here. This is horrendous and looks like what most A.I. art looked like 4 years ago.

While it looks somewhat better far away, wimmelbilds are not made for the 2 second glance to then just scroll past which is probably why this type of art doesn’t work well on short form social media platforms. Wimmelbilds are meant to be studied and looked at and savored, but here, further inspection just leads to disappointment and confusion. The prompters are missing the point here, as none of the images are even fun to look at!

So, while I believe there could be improvement with a model trained specifically on wimmelbilds to create better images, the ever-present question of “but why though?“ still lingers. We are delighted by wimmelbilds not just because they are super busy pictures but because we are amazed a person made the image and did so with skill, knowledge, and humanity. You can laugh at all the little stories and humor the relatable human artist packs in and you can revel in the detail the human artist dedicates themselves to and includes or get the pop culture references hidden inside it, its the humanity that makes it interesting. I just don’t think A.I. needs to take up this space.

When using A.I. , you are just a customer to the companies promising the democratization of art, art for those who can’t art, or whatever edge case uses for this technology you can come up with in the art space. Maybe you are a customer, one who will “know a good result when you see it“ or maybe you don’t fully know what a good result is and you will unknowingly force everyone else to slow down, look closer, and just to see how bad it really is up close.

~DM_Polymorph