/ic/ - Artwork/Critique

>>6858798
>You say this now, but in another reply to another anon you say that IA can only work based on what it has been given in data. Which is it then?
you're confused..
i'm saying that is using the data to create something new. or at least new according to what it has learned.
so let's say there is no image of a quokka drinking a coffee in its training data. it can still create that image because all the parts are in the training data, allowing it to work with all the concepts separately.
because again, it does not take your images as they are, it takes the concepts and ieas and applies those to the new image.

>but the reality is that the bast majority of the people only abuses this and only want to replicate what is already done in popular trends or a prexisting style that is recognizable with ease, despite that it can create something "completely new" according to you.
what does it matter what the trends are now. in my opinion it's not much different from how it was before. even good artists were a dime a dozen.
and this is a different discussion about the worth and impact of AI. i don't want to get too sidetracked.

>Hell, even the shit you have been posting looks totally void of personality despite the superfitial technical competence of what the AI is doing. For most normies it may be enough, but for most of the people on this gig, its unappealing
sorry but that's just cope. even from your images, don't you find the astronaut image obviously appealing?
and it is also not wise to think that AI is limited in terms of style. not when it can create photorealism well enough to create deepfakes.

and i'm not using the term cope as some sort of meme insult.
i'm saying that you saying these things
>AI will always be [insert anything]
>AI will always be unappealing
is literally the artists way of coping with the coming changes. it's just wishful thinking.

>>6852445
that's the only point you CAN make, you'll realize. because it's so broad it will cover just about any action that takes in data in the universe.

including our brains. we take data in, adjust our neurons, and do what we do.

let me ask you another question: how do you think YOU would try to draw a catdog? in terms of what is going on in your mind?

>>6852451
i have before. a really blurry version of it.
i posted cropped sections of a sketch before as well. though not under this trip iirc.
you know what anons said at the time? "i couldn't prove that i drew that".
this should tell you everything you need to know.

>>6846983
>an AI can instantly present and recall any graphic it "learned".
it can't, and that's not how it works. researchers have actively tried. and iirc they came up with a rate of 0.03% or something. so 3 out of 10 000. it is exceedingly unlikely for it to produce its training images. and that's when TRYING, in normal use it's pretty much impossible nowadays, especially with finetunes & lora.
that's because the training data is GONE. all of it has been converted into what it has "learned", which is:
>how to turn noise into imagery that conform to keywords (your prompt.)

>Humans don't "create a statistical model based on every picture they've seen in their life".
they kind of do. definitely not "statistical", but neither does AI. the actual maths is beyond you and me.

when arguing this, you have to keep in mind that a human will have far, FAR more datapoints to grasp from thanks to real life vision. you can see how a cat moves and turns, allowing you to see ALL its angles and how it moves. you can even touch it and stuff.
- but if you only had one single photo of a cat from the front you'd not be able to draw it from the back. (you can make up some stuff by inferring from other mammals, but the AI could do that too, if it was tagged this well, which it isn't.)
- if you had two images of a cat, then you'd be able to draw quite a bit more. but it would still be limited.
- if you had a hundred high quality images of cats in very varied situations, you would do significantly better.
what i'm trying to say is that we do infer from what we see. we're not "averaging" things, but we can only use what we do know as datapoints for our own "inference". The same goes for AI. but it has less data and is less smart

>they have to learn the actual construction of their vision.
AI has to construct things as well. just in a different way. it does shape things out of noise. and things slowly form as the image denoises. first the rough composition, then shapes and then details