Why so much hate toward AI art?

[Pr0GamerJohnny]

Imagine watching your favorite anime but with daz art.

I'm not convinced, sometimes bad art can very easily tarnish a great text-based story, instead of allowing readers to fully imagine the scene the way they like you are now making them look at low effort art.

Don't make the mistake of conflating ai art = closer to the idea of daz 3d stuff than hand drawn cartoon stuff.

Those two aren't opposites, both can either be a creative endeavour or from scratch - the analogy for handdrawn would be freehand vs stencils etc.


In other words, creating stuff in Daz and similar styles isn't the enemy. The enemy is mass-produced, uninspired art which can certainly occur with 3d models (COUGH WVM) or cartoon-styles (COUGH ORS).

 

[d7anime]

nah AI art is going to take over. ControlNET for StableDiffusion is here and Im definitely going to use it and chatGPT to make a free game.

 

[Hu3hu3hu3]

The only people mad are those who are losing their business model because AI does a faster, better and cheaper job ¯\_(ツ)_/¯

...and it will only get better.

 

[subatomicweasel]

Sorry for reviving a dead thread, but I figured that I'd throw my two cents in for it. I'm generally in favor of AI generated art, and I use it quite a bit for myself. However, I put a huge asterisk next to the first part of that statement, and I do think that there's an ethical line. If it's for personal use, go ham. I personally love to create images in the style of Boris Vallejo and Frank Frazetta with my own ideas because I love that style. However, if you use it commercially to pass something off as your original creation, then that's where I see a problem because the AI program is just remixing other peoples' work. I personally use it for character concepts and for getting past writer's block, and I use AI Dungeon to help practice my writing.

With the recent US ruling on whether or not it can be copyrighted, I think that it'll be reduced to a niche once the hype dies down and most of the user base will be reduced to us degnerates. I'd use AI Art for a good wank, but I have yet to find a solid one that does hardcore instead of just softcore porn (outside of PornPen) that doesn't require me to install a bunch of shit onto my computer. I've even looked for subscription services and pretty much screamed "take my money" and still can't find anything, so obviously AI generated art has a long way to go.

 

[Winterfire]

the AI program is just remixing other peoples' work.

Except that's not how it works.
I will never understand why people always get this point wrong when google is accessible to everybody and it takes only a few minutes of search to know how something works.

I will give that you can make remixed work with AI (Which are the only examples ever used for arguments against AI art), but that's just one of the many uses.

 

[Laikhent]

This was actually generated by an AI (midjourney5). Crazy isn't it?

 

[benisfug]

You could apply most arguments against AI art to most 3dcg too. Worst that's going to happen is you having to filter another tag.

 

[woody554]

Except that's not how it works.

after 6 years of minoring on information theory and neural nets I'm opinioning that it's exacty how it works.

all neural nets are classifiers that divide inputs into n-dimensional space where n is the number of 'correct outputs' it's been trained to 'recognize' ie categorize. what the new batch of nnets are doing is making linear combinations of those, ie. LITERALLY "remixing other people's work" and nothing else. there is absolutely no 'thinking' or 'creating' going on, just copying and interpolating within the vast mountain of training data created by actual artists.

now, it could be claimed that our brain is doing the exact same thing, and partly that would be true. except that it isn't, and our brain is doing something weird called 'thinking' in addition to it, and we have no clue whatsoever where that 'thinking' part comes from. (it's not 'soul' though). even the aforementioned basic model of how nnets classify we kinda ran into by accident, and we don't really know what happens to the weights inside the nnet.

I mean we know of course how we're adjusting the weights, but we don't understand the emergent content of it. in other words, we can't 'read' the 'thought' from the neurons even though we know exactly their state and how we adjusted them. we just treat it as a 'black box' and wait for the output at the other end.

one day we'll definitely crack it, because it's no magic and there's no 'soul' involved in any of it. but today we don't really have a clue where to even begin. people are just fooled by the pretty pictures and language that LOOKS natural, because they don't understand there is zero cognitive processing behind it. chatGPT is not 'thinking' when is spews out that 'speech', it's mechanically combining the texts of OTHER people and generating an output that LOOKS LIKE the linear combination of its training data.

but I wanna repeat that this does not mean 'machines will never be sentient, they just copy things very well'. you know, what data was said to do in star trek. it's just that we're not there, and we have no idea what we should even try to start the journey there. right now it IS just clever mindless copying, but one day will figure out how to make nnets actually think. so that they can create new stuff and not just copy others.

 

[Winterfire]

after 6 years of minoring on information theory and neural nets I'm opinioning that it's exacty how it works.

all neural nets are classifiers that divide inputs into n-dimensional space where n is the number of 'correct outputs' it's been trained to 'recognize' ie categorize. what the new batch of nnets are doing is making linear combinations of those, ie. LITERALLY "remixing other people's work" and nothing else. there is absolutely no 'thinking' or 'creating' going on, just copying and interpolating within the vast mountain of training data created by actual artists.

now, it could be claimed that our brain is doing the exact same thing, and partly that would be true. except that it isn't, and our brain is doing something weird called 'thinking' in addition to it, and we have no clue whatsoever where that 'thinking' part comes from. (it's not 'soul' though). even the aforementioned basic model of how nnets classify we kinda ran into by accident, and we don't really know what happens to the weights inside the nnet.

I mean we know of course how we're adjusting the weights, but we don't understand the emergent content of it. in other words, we can't 'read' the 'thought' from the neurons even though we know exactly their state and how we adjusted them. we just treat it as a 'black box' and wait for the output at the other end.

one day we'll definitely crack it, because it's no magic and there's no 'soul' involved in any of it. but today we don't really have a clue where to even begin. people are just fooled by the pretty pictures and language that LOOKS natural, because they don't understand there is zero cognitive processing behind it. chatGPT is not 'thinking' when is spews out that 'speech', it's mechanically combining the texts of OTHER people and generating an output that LOOKS LIKE the linear combination of its training data.

but I wanna repeat that this does not mean 'machines will never be sentient, they just copy things very well'. you know, what data was said to do in star trek. it's just that we're not there, and we have no idea what we should even try to start the journey there. right now it IS just clever mindless copying, but one day will figure out how to make nnets actually think. so that they can create new stuff and not just copy others.

Images are not stored anywhere.
The pieces created are unique, and are not a remix (if by remix you mean the AI taking a piece off an image, then another piece from another image like it was making a puzzle).

 

[Pr0GamerJohnny]

Images are not stored anywhere.
The pieces created are unique, and are not a remix (if by remix you mean the AI taking a piece off an image, then another piece from another image like it was making a puzzle).

I mean by that logic any rap song that samples something is unique, since it merely took bits and pieces from existing art.

Of course, that's hogwash, so all we're really debating is the "resolution" of the sampling - whether broad trends and groups of data and pixels are used or individual values and neighbor's values. To me, it doesnt really matter, it's all derivative; and not unique.

That doesn't mean it can't produce works that SEEM unique, but that's because our eyes are looking at the broad picture, not the digital footprint. Also, this doesn't mean it has no purpose, it certainly does, but dont try and pass it off as something unique or creative.

 

[Winterfire]

I mean by that logic any rap song that samples something is unique, since it merely took bits and pieces from existing art.

but it doesnt

 

[hakarlman]

What the AI scene needs is a list of tried and true, reproducible methodologies that can be used in a production pipeline.

Example:
1. A way to reliably change the facial expression of a person in ANY image; preferably without having to upload the image to a remote server, but using offline stable diffusion.
2. A way to reliably make clothing tighter; offline stable diffusion needs to reliably detect all clothing with precision.
3. A way to reliably split every object in an image into sub-objects that can then be removed or changed.
4. A way to turn any object in an image into a re-usable 3D asset/representation; this will need 'training' I think it's called, but they need to make it easier to do.

AI is going to make it into nearly every facet of digital media IMO; especially porn, it's coming!

 

[baloneysammich]

I wonder how companies that had begun to integrate AI art into their workflow are reacting to the copyright ruling. I was on the Stardock forums a while back and the founder (IIRC) said one of their next games was gonna use AI art pretty heavily.

 

[Winterfire]

I wonder how companies that had begun to integrate AI art into their workflow are reacting to the copyright ruling. I was on the Stardock forums a while back and the founder (IIRC) said one of their next games was gonna use AI art pretty heavily.

I might be wrong on this, but I believe the copyright ruling only applies to text to img as-is.
There are many ways to use AI, you can even do the outline yourself and let AI color it. The outline would still be theirs.
Alternatively, you could generate 100 images and use them to do photobashing (which would otherwise be done with other images), blender has even a plugin which lets you generate AI textures... You will hardly use them as they are, especially since a game-ready model wouldn't have multiple materials, so you may as well edit them slightly on your favourite image editor.
Even Photoshop has AI.

There are tons of uses and softwares using AI already, but people who cry about "but muh artist!" only ever use 1 or 2 examples of people who take someone's image and run it through img2img.

 

[Meaning Less]

I might be wrong on this, but I believe the copyright ruling only applies to text to img as-is.

It isn't just that, any software TRAINED with previously copyrighted material can fall under this new ruling depending on the results of

You must be registered to see the links

.

 

[Deleted member 2741424]

I actually love it. It's a great tool, and I can't wait to see how it's improved in a few years.

 

[Winterfire]

It isn't just that, any software TRAINED with previously copyrighted material can fall under this new ruling depending on the results of

You must be registered to see the links

.

I am not a law expert (or knowledgeable at all) hence why I wasn't sure, although I have to wonder what happens if it is transformative. I mean, if you can do photobashing with photos and have it hold copyright because it's transformative, I can't imagine how AI wouldn't also count as transformative when it doesn't use images at all.

 

[Meaning Less]

I can't imagine how AI wouldn't also count as transformative when it doesn't use images at all.

The AIs we are talking about here do use images though, that's how they are trained, they do not work unless you feed them with tons of training material.
And that's what this ruling is going to be about, how ethical is it to train those softwares with copyrighted material and what limitations should apply.

This video has some interesting information on the topic:

 

[Winterfire]

The AIs we are talking about here do use images though, that's how they are trained, they do not work unless you feed them with tons of training material.
And that's what this ruling is going to be about, how ethical is it to train those softwares with copyrighted material and what limitations should apply.

This video has some interesting information on the topic:

Yeah but the important bit is that they do not store or remix/use those images, this is something most people get wrong. Not sure if that changes anything when it comes to copyright though.

 

[Meaning Less]

Yeah but the important bit is that they do not store or remix/use those images, this is something most people get wrong. Not sure if that changes anything when it comes to copyright though.

It does do exactly that though, it just does it their own way.

There is zero intelligence behind what it does, it just scours through all the training data to find the closest inputs to what it was requested of it, but ask anything out of their training dataset and suddenly it is completely lost.

Like here I used the

You must be registered to see the links

, which probably isn't trained with many animals and asked for a simple "duck", here is what you get: