That’s an interesting topic - I feel earlier algorithms were able to give you more “emergent” and chaotic results due to being less tuned and refined - with some experimenting you could get something - not “finished” but certainly unique-looking. Nowadays they’re very much tuned to produce very finished-looking artworks that are also very samey and “median”. They do look competent but have that unmistakable “if you seen one you’ve seen them all” feel for any particular style you add to the prompt.
Metaphor…I could have went off and said, take away my brain and I can’t function x) while some can draw with their feet it does not really add anything to the dicussion, hats off to the talented people ofc.
Though what I meant was, there’s a difference between making an exact copy of an art and using thousands of imaegs as references to as you put it, takes bits from thousands and makes an amalgamation of it that doesn’t look like any one thing.
Mhm, I seen some with AI arts in sw on their trps, they feel quite samey, again it can’t replace real arts.
Time is the real issue I believe. I can makemy own rbead if I try hard enough but I would rather buy some, if my economical situation isn’t good, likely from the cheapest source.
Drawing is a skill, almost anyone canq learn, albeit time and effort it takes varies.
A computer or an AI is not a person, it does not have the capacity of original thought. AI is a predictive system built on patterns it ‘sees’. It cannot by definition generate anything original ever.
It simply reads the prompt and predicts what the result should be, what pixels next to existing pixels should look like when it’s ‘generating’ art, not thinking organically like an artist trying to choose the correct colour or shade to whatever you’re drawing.
It’s how things like chatbots and phone’s word suggestions work - it reads your word and predicts what the next one should be based on patterns the algorithm has been fed. Giving the AI more specific prompts limits their data and narrows all the data down, so if the AI writes a story it can be of anything, and when you specify you want it to write a story about specific classified documents, it’ll write up those documents. If you want a certain artist’s look, it’ll ‘generate’ something that falls into the patterns it has learnt from that certain artist’s look.
A human person - let’s say an artist to be on topic - can see another person’s artwork and think ‘wow, this is REALLY cool, I want to do this!’ and will not be able to immediately remember everything about how the work was composed, drawn, etcetera. If there’s a tutorial (f.e. the artist has made a tutorial on how to draw the work) they can follow it, perhaps learn a technique used to achieve a goal. But you cannot really organically come up with the same work yourself, especially if it’s way more detailed than a simple two-colour comic-style drawing of someone.
Now that you’ve learn the technique, you can implement that technique in your own work for other things. You’ve actually learnt it. The AI however simply predicts what the pixels should look like in that style. If the artist has a really unique style - let’s say Pointillism like Seurat and Signac - you can learn from them and if they were alive today, maybe teach you how to properly use that style, but a program will simply predict what each pixel now should look like given the prompt.
The problem is - for the AI to ever have the possibility of generating art (or text, music, so on) it has to be fed data. If you’ve only fed AI data of let’s say Acrona’s art, it will not know what to do with the prompt if you ask it to draw in the style of pointillism. To be as effective as it is right now, it needs A TON of data, and realistically you will never be able to not even just get permission for, but even just reach out to all sources of data to get the amount that has been gathered thus far.
So, the creators of those AIs have simply… stolen them, without permission and not even asking, by scraping the internet for all the data it needs. And the scraping is done by a machine as well, you don’t have a human going through ArtStation profiles and downloading everything, it’s an automated process, because you cannot easily give the computers usable data that a human processes immediately.
Maybe a good metaphor is basically telling someone who’s born blind what to draw. Telling them to draw in white and blue means nothing. Telling them to draw the webpage of Walmart means nothing. Telling them how much to move their arms where, or where to get proper colours from gets you somewhere. The blind person will still end up predicting where to move their arms, fingers, body to get the result you’re looking for, based on what you’ve told them. And thus you end up ‘training’ the blind person how to draw a Walmart webpage, and the blind person with copyright infringement.
The computer can’t do anything on their own though, but the human has agency and thought and can decide to do with your information as they will.
6 Likes
There is - but the issue most people have with it isn’t direct plagiarism of final images, but the fact that artwork used to power the AI - it wouldn’t have been able to generate fantasy digital art without being fed fantasy digital art first - was sourced for commercial usage in the AI without consent or licensing.
Let’s go for a gradual comparision:
Imagine if your artwork got yoinked for use in some “Fantasy Art Clipart Collection”, sold online without attribution or a single penny going your way.
Now, what if that collection had a handy search function - would that change the issue?
Now, what if it had some kind of remix algorithm mixing artwork together, based on your search terms?
What if there was a really great amount of artwork, and the algorithm really advanced, enough to legally obfuscate you from directly pointing your finger at the result and saying “hey that’s mine”?
I feel that’s kinda what it’s like.
2 Likes
Hrm…I see the logic in what you say, but stealing some bits from Nerathion’s bit here. So AI is fed material which it uses to pedict which pixel goes to where when asked for a specific promt?
AI isn’t self consiouss and lacks creativity, that ‘spark’ if you will…Unlike a human who also uses other arts to learn, to predict and draw lines and know where they are supposed to be. But humans have tha sparks that makes it so we add our own…flair? as we draw.
Though on contrast to AI humans can’t perfectly remember thousands of arts and process them mentally.
So…both works in different ways. One is a cold machine that takes a larger data base and uses it to make something without any sort of creativity. Yet a far larger sum data that makes the end product different than teh thousandsof parts it used as reference.
And on the other side you have a human who uses a far smaller data pool but makes up for the difference in talent, creativity and flair, creating a far more unique and special piece: ie an art.
TBH whole things sounds like somehing out of a sci-fi movie (hope I do not sound old with that). Though I suppose Iam partial, this whole discussion made me question the definition of words such as creating, learning, etc. Thank you all for the feedack and genuinly well thought through answers.
2 Likes
Another thing to say about this data pool, which, sorry for repeating myself from a few posts prior; it’s been revealed that a lot of this data wasn’t just taken from other people’s art. In the case of some AI engines, it was taken from over a billion photos of real people on Facebook, some of which including children, loved ones long since deceased, and so on.
It’s something straight out of some Cyberpunk Dystopia because it… sort of is pretty dystopic. It’s sort of a reverse Wall-E situation where robots are being ‘trained’ to ‘do art and write’ while humans are still doing minimum wage jobs.
Dystopic as it is sadly, leakage of personal data from social site already happens, of course it is wrong in any shape or form.
Like all tech I think AI can be good or bad, recognizing the face of wanted folks etc could be a boon for instance as for the rest of the comment.
Learning art isn’t as simple as ‘picking up a pencil’ sadly…I practice, I am still quite bad but I keep at it as I like drawing, but point is, it isn’t as simple for most people. I rather hire an artist to draw x) if it is for a genuine long term character.
If people are using it to make profit that’s quite dark though.
The bad part is that it’s the actual company that is running Facebook that did it, so, not a leak in any regard on that one case.
And, most AI processing engines cost a monthly subscription nowadays, if I’m not misremembering.
I wouldn’t know, though, I was more referring to people using AI stuff and say they drew it and sell it as genuine ‘art’.
Company still ‘creates’ something by writing algorithms, ergo, is selling a product, in this case the ai.
People who in their sick and twisted way make and sell ai ‘art’ that I cannot fathom.
I would moreso give a different example. Let’s say you want your character to have armour. An artist can imagine (or look at references) for where the armour goes. AI simply looks at the data of characters with armour and generate a similar result.
You decide you want to give your character a certain pose. The AI would look at data sets of poses and generate something that resembles said pose. I don’t really know how the more advanced AIs work - the machine probably has some internal logging of its generated output of what should be classified as what. It can detect changes in colour or surroundings to detect whether a group of pixels should be regarded as ‘arm’. Therefore to resemble that pose it’ll try to redraw the first output with its internal logging in mind. If the character was naked, I’d expect the AIs right now to reasonably well draw the same character in different poses.
The armour isn’t one solid block of material, however. Maybe it has chains that connect to each other, and intricate detailing on moving parts. The artist can visualise how the armour changes and adjust the picture accordingly. The AI, probably in the future, will be able to keep track of detailing as well, the developers have already managed to fix the early tells of AI generated imagery (hands were a very easy sign - the AIs simply couldn’t come up with correct predictions of what the positions of fingers should’ve looked like… ironically, hands are often regarded as very difficult to draw for many artists as well).
However, to even generate any detailing for the armour parts, the AI generates what it thinks detailed armour looks like. There’s usually a lot of fine work, and the artist has the edge right now- they know where things should be long and straight lines, what should connect how, what needs to be separate, etcetera. They can draw helpful lines or plan them out on their own to make sure that one particular detail has continuity. The AI starts falling apart.
Right now at least, when you google for AI images and actually look closely, you can tell very quickly that it’s wrong. Of course, not every artist wants to painstakingly follow extremely detailed things to make sure they’re all correct in the end - for example, if you’re drawing a woven basket, I can imagine the artist taking shortcuts so that they wouldn’t have to literally draw every single woven element, or in woven fabric every thread. On armour, let’s say chainmail, an artist may simply do short round squiggles to give the impression of chainmail. AI might do the same, but… hey, some of those squiggles are really long like spaghetti instead.
AI needs to have a LOT of data to be able to train itself to deal with those details. You cannot simply ‘teach’ the AI how to draw fingers, details, or armor. You cannot talk to it, or write code that for chainmail parts each chain has to be drawn this way. What if the chainmail armor is damaged, or warped? A human can visualise those changes, a computer needs a human to visualise it for it. The more data AI has, the more it can detect those patterns to fine-tune how to draw more detailed works. With far more data on what fingers look like, it can predict where fingers go. But it cannot visualise nor think, or properly review its ‘work’, as the more it tries to be detailed the more like a particular piece it’ll look.
For example, some artists (and writers) have been able to detect certain patterns in the AI’s generated output and gotten the unofficial confirmation that indeed, their works were stolen. The AI spew some aspects of the stolen work back at them, as the more specific prompts you give the narrower the data sets become.
I’ve done exactly this for my night elf, because while I can pull off an astounding Worgen, I cannot really get a comfortable voice fit for a Nelf.
1 Like
Since I promised I would, I will present two examples of results from tools to use instead of AI generated art!
First one I made quite a while ago of a halfling druid I was playing in a D&D campaign. It was done using a doll-dress up/avatar creator, known as “Rinmaru’s Mega Fantasy Avatar Creator”. Sadly the original version does not work anymore because flash, but there are uploads of it elsewhere, and other alternatives to it, but it’s an example!
The second is Heroforge!
Now, you can do alot more and better than this on Heroforge. Both of these were done today only a little while ago and mostly as an example, so I rushed it a little. With the latest feature having proper sculpting, you can spend a few hours working on a good model. And it’s free to use, save and screenshot unless you wish to specifically buy the model for printing(or upload into virtual tabletop).
But yes, just two examples of things you can use instead of ai generated art if you feel like you can’t draw yourself & or lack either the funds or willingness to hire an artist.
12 Likes
Heroforge is so much fun to play around with and is a great alternative to using AI art to make reference for characters. It’s really versatile 
4 Likes
I did so in the past for Valteryl specifically, and so did some other friends too. I do, however, know that the person stopped doing commissions as a freelancer and was employed by a studio as a full-time VA due to simply “not having enough work due to the rise of the AI VAs”.
1 Like
Unfortunately heroforge is crap if you play a bestial race.
I know they have minotaur models but it’s not the same.
It is what it is.
4 Likes
I’ve considered it, but don’t know how to contact the VA I had in mind 
I also want to commission an artist to create a theme song unique to my character. I knew someone who did them a while back, but has since then stopped to focus on their personal projects.
It really is! Especially within the last year or so they’ve added so many more options and ways to customize the models even further. If Heroforge wasn’t a rabbithole in the past, it definitely is now. You can sink hours and hours into customizing your character.
I’ve been tempted to order the actual 3D figures from them. Would be cute to have a character or two sitting in my bookshelf 
1 Like
I’ve not bought one myself yet, but for the services they offer and types of quality for the models, their pricing isn’t actually that bad.
1 Like