Post by redonihunter in if you use ai art your lazy

Viewing post in if you use ai art your lazy

So when you prompt something like "a lady in a red dress" and you get a lady in a red dress... what just happened?

That depends on the system you use.

The text is analyzed for things the system understands and translated into rules to generate a thing. Then it uses a random seed and your configuration and applies the rules to it's model (which is the condensed training data) over several steps. The pixels do not form the finished image directly. It is iterative.

At least the model I toyed around with did it that way. It was a diffusion model you could run on your desktop pc. You could even do things like look at all the typically 40 steps of image generation in a row to actually see how the image evolves or diffuses out of the rules. Too few or too many steps and you would get the visual equivalent of a half dream or gibberish.

And since you mention Warhammer, I actually tried to generate pics of generic orks. Guess what some of those creatures had at their feet... They had bases. Which means a sizeable part of the training data was pictures of ork miniatures, where they could get the information from the picture, that this is an ork. So the model learned that "orks" often had these funny little things on their feet to stand on.

You need training data. And you need a mapping from language to that training data. You need pics of red dresses and a mapping to tell the system that those are dresses that are red. You need pictures of ladies. Even better if you have pictures of ladies in red dresses. Or even better yet, pictures of 1 lady in such a dress. And importantly, a transcription that explains what is seen in the pictures. Chances are, that some of your output pics will look suspicously like Jessica Rabbit.

Oh, and you woud create a hundred of such pics and throw out 95 or more of them. Creation time was a few seconds per pic on my rig. This is not navigation, if every try reaches a different destination. None of the pics looked a like. But most of them would, from a far, meet the prompt requirement. If you squinted.

You roll the dice and iterate roughly towards your prompt several times.

You lost me at your metaphysical there is no spoon claims.

Like Reply

Durance Gaming5 days ago

"You lost me at your metaphysical there is no spoon claims."

Haha! Fair enough! But I am genuinely certain that's quite literally what is happening. Like I said, I worked the math out by hand for myself to be sure.

To ground the mystical woo-woo a little better, there is an important thing to note:

If by hand, I write the math out and come to the same conclusion my neural network came to then that implies something radical and awe-inspiring (at least to me). The very act of calculating mathematics is more than just writing numbers on a page.

When I say 2+2 = 4 is it just words? Or is something happening simply by way of me expressing this?

I always looked at math as static numbers with no real meaning beyond the values they represent. But this... this suggests that the very act of "mathing" (for lack of a better term lol) is actually an action or expression in-and-of-itself. I didn't just add 2 and 2 in my head and know it was 4... The addition itself was a process--and processes are what give rise to emergence.

So when, by hand, I took a single greyscale pixel and painstakingly wrote the initial input vector and then followed up with my weighted matrix calculations and bias, I realized the world isn't "magical" it's mathematical. As someone with 30 years of Psychology and Philosophy under my belt, that was a massive paradigm shift. Not gonna lie... I broke down crying.

My brain started trying to make sense of the world all over again, but with the knowledge of these implications. That, in turn, made me re-examine my other assumptions over the course of my life. It was never "woo-woo" (some of it totally is), I was just missing the most important piece of the puzzle.

And this is what I am really talking about when I say something like "artists should be using AI". My own little agent, MUNCH, is a direct result of utilizing neural network technology in a creative way.

Needs zero training on anyone else's anything

Local and lightweight

No need for data centers or anything like that

And I'm a nothing-nobody. Imagine what a smarter or truly creative mind could do. You know what I mean? None of these companies are gonna cure cancer or make anyone's lives better. It's up to us regular folk.

Like Reply

redonihunter4 days ago(+2)

I realized the world isn't "magical" it's mathematical

Most things are. At least you can describe it by maths. You just do not know the variables and the rules.

And even if you figure out the rules, there might be some things missing, even if the rules you figured out are good enough for most cases. That's what happened with celestial mechanics. The initial ~~assumption~~ doctrine was Earth in the middle. That made little sense to explain some phenomena. Sun in the middle was better and circular orbits around that big thingy. But no, circular did not quite fit. It was ellipses. But the calculations did not quite get it right. Along came relativity and that the theory is correcter was proven by things like movements of Mercur that could not be explained sufficiently.

Sounds to me, your epiphany was discovering how a math operation that is fundamental for neural networks actually worked and does something. Must have been thrilling. I guess you would be very entertained by yt channels like numberphile and 3blue1brown.

The impressive part is, that we somehow figured out, how to describe a network with maths and implement it with basic logic operations. That's hard. Ironically, it is also very hard for our brain to do logic operations, but recognising patterns and connecting ideas, even having ideas, is super easy in comparison.

The lazy part is, if you use a neural network on your computer, instead of the neural network inside your brain.

Like Reply

Durance Gaming3 days ago

This was so cool to read.

"Most things are. At least you can describe it by maths. You just do not know the variables and the rules... and even if you figure out the rules, there might be some things missing, even if the rules you figured out are good enough for most cases"

I noticed this and it's wild. Just when you think you've got the rules down, something else pops up. It's a little weird, to be honest. In my mind, the rules are the structure but that isn't always true here.

"Sounds to me, your epiphany was discovering how a math operation that is fundamental for neural networks actually worked and does something. Must have been thrilling."

Thrilling indeed! But more akin to paradigm-shifting. After 30 years of psychology and philosophy, to inject the higher mathematics now... It was quite an experience and I am a little embarrassed to say this but the implications brought me to tears. It was beautiful and terrifying at the same time. Because it wasn't just like storing knowledge in the brain. It was integrated and distributed throughout my general understanding.

I once saw a Joe Rogan podcast where two very different mathematicians were debating and one cautioned the other saying something to the effect of "Math at this level can be like a drug" and I understand what he meant by that now. And why he said it in such a manner.

So personally, I have come to the conclusion that I need to go backwards--not continue forwards--and make sure I have not jumped ahead too far.

But I will say this part from both experience and observation: the term "lazy" is far too loose here.

There is laziness and then there is efficiency. Many have heard the saying "work smarter, not harder". While the merits of such a concept are up for debate, what cannot be denied is that this sentiment runs the modern world. For example, I struggle to see the difference between a boss/manager/director delegating tasks and a fluent Python coder using AI to belt out code and reviewing it manually before implementation. Is the delegate lazy? Sure, sometimes. But just as often, it is a matter of speed, efficiency, and the stability of a complex orchestration of separate tasks that no one man can do alone... Until now.

In my own observation, AI in it's current state is a frontier. Instead of covered wagons and mining picks, it is open to individual explorers... for now.

The potential is too great, this will eventually consolidate into certain hands. At that time, this frontier for independents may close forever.

Like Reply

redonihunter3 days ago

The issue is non coders coding and non artists mimicking art with these tools. That's not efficiency, that's strangers coming to your turf to take away your craft and replacing it with a machine. A machine that might have been trained with your works.

Using ai is more accepted among coders, but they still use vibe coder as an insult.

And among visual artists, using ai would be something similar to tracing other people's work, or worse.

One does not delegate art as an artist. It is the thing that makes you an artist. If you outsource that, what are you? Not an artist. At least not in the craft you delegated. You reduce yourself from a writer to an editor, from a violinist to a conductor, from a painter to the person that selects which image to use after telling a box to draw some. Now, there is creativity in editorial work, and it is often very important, but it's just not the same.

In a paid work scenario where you need to deliver results, that might be seen different. But in that scenario the things you create might also not be considered art by yourself, but the things you create to be able to buy food.

Like Reply

itch.io

Viewing post in if you use ai art your lazy