WotC WotC: 'Artists Must Refrain From Using AI Art Generation'

WotC to update artist guidelines moving forward.

After it was revealed this week that one of the artists for Bigby Presents: Glory of the Giants used artificial intelligence as part of their process when creating some of the book's images, Wizards of the Coast has made a short statement via the D&D Beyond Twitter (X?) account.

The statement is in image format, so I've transcribed it below.

Today we became aware that an artist used AI to create artwork for the upcoming book, Bigby Presents: Glory of the Giants. We have worked with this artist since 2014 and he's put years of work into book we all love. While we weren't aware of the artist's choice to use AI in the creation process for these commissioned pieces, we have discussed with him, and he will not use AI for Wizards' work moving forward. We are revising our process and updating our artist guidelines to make clear that artists must refrain from using AI art generation as part of their art creation process for developing D&D art.


-Wizards of the Coast​


F2zfSUUXkAEx31Q.png


Ilya Shkipin, the artist in question, talked about AI's part in his process during the week, but has since deleted those posts.

There is recent controversy on whether these illustrations I made were ai generated. AI was used in the process to generate certain details or polish and editing. To shine some light on the process I'm attaching earlier versions of the illustrations before ai had been applied to enhance details. As you can see a lot of painted elements were enhanced with ai rather than generated from ground up.

-Ilya Shlipin​

 

log in or register to remove this ad

And I'd say otherwise. And thus we distill the debate down to its basics. That's what it all hinges on.

Indeed, I dispute the use of the word 'creates'. AI doesn't create. It literally can't.

Let's say "generates" then. The CJUE (nor the original German court that decided that it was OK to sample in the case of 2 second sample) doesn't consider how the sampling is done or whether sampling is art or not. It considers the question, fundamental in IP: is the product being distributed by the defendant (irrespective of how it is made) copying the totality of an IP-protected product, or a recognizable part of the plaintiff's existing IP-protected work. Can someone look at a product and say "I recognize this work of art in this product"? If yes, then it's forbidden (therefore, sampling long parts of musical pieces is forbidden unless one acquires the right to the original artwork), if no, then it is legal and the sampler could use it without IP-law problem.

The distinction you make is irrelevant to the debate over potential copyright infringement made by the one who distributes the generated image. It is however extremely relevant to whether the distributed product can enjoy IP protection in itself. A court might rule that anything randomly generated can't be copyrighted by lack of intent, another might rule that the person running the generation will have created the thing in his mind and just used a tool to make a graphical representation of the image he has in his mind. A third court (or national juridiction) could decide that an IA can't create and therefore what it spurts out is public domain, but the transformations made to it before publication by a person are enough to make the end result IP-protected... It's a very interesting legal debate, but the outcome would be whether the generated product can be copyrighted or not, which is a different legal question. I am not sure having the generated product instantly be public domain would hamper the development of generative IA, but this is difficult to foresee (I'd say it would be problematic if one wanted to create a company's logo, but it wouldn't be detrimental to people who use it for illustrative purpose in an otherwise copyrighted work).

Whether it is ethical is another debate -- one that will be more difficult to solve as some societies might want to go more toward defending the individual rights of the copyrights holders, others will consider that the need of inexpensive, average quality art for all outweighs the need of the few enough that an exception can be made to the temporary monopoly usually granted to them, other will be happy with a solution where a special tax is created on profit made from IA as a proper compensation, and other might make art generation a public service offered in public libraries... lots of ethical solutions can be imagined, and the possibility of unified outcome are much lower than the "more or less unified" field of IP laws in the world)
 
Last edited:

log in or register to remove this ad

J.Quondam

CR 1/8
Note that it will also have about a bazillion references to unicorns as "horse with a horn", which means all the examples of horses will also apply. And horses are quadrupedal animals, so those references apply...

Which all means that the volume of text references that lead you to somethign approximating a unicorn is probably larger than we expect. Unicorns are not obscure.
Yes, exactly. We already know an LLM "understands" the relation between horses, quadripeds, horns and all that based on the huge amount training text. The feat is getting from that internal text-based representation of "unicorniness" to an image, without ever training on images.

Near as I can imagine, it's like reading a bunch of material about unicorns - and all relevant adjacent topics - to a blind person, then having that blind person draw a unicorn.


It would be interesting to see what happens if you try to get it to draw something more obscure. Like a squonk.
Probably nothing more interesting than asking a human to draw a squonk, I suppose?
So I asked Chat GPT about "squonk". Turns out it's a type of e-cigarette. Then I clarified I wanted the creature, so it talked about some cryptid in Pennsylvania. Then I asked it draw an ASCII image on one, and it drew...well, something, which I'm posting only because it made me chuckle:

Me: could you draw a squonk using ascii characters?
ChatGPT: Of course! Here's an attempt at representing a Squonk using ASCII characters:
Code:
python

                    .-'''-.
                   /       \
                   |__   __|
                    |  |  |
                   .-'  '-.
                  /_________\
                .:_:_.-:_:_:.
               | o   (_)   o |
               |    _______  |
               | .-'  12 `-.|
               |/   8  _  3  \
              |\_._     ___..-\
        _.--':  |  |'''      | '---._
       /   _   \.-'\          /.-'   `\
       \____\____\  \_ ___ ___/ /______/
        `. 8 8   |  \________/  | 8 8 .'
          `.__.__/  |  __  __  | '__.'
            `-.__.-'   `'  `'  `-.
Please note that creating detailed ASCII art can be quite challenging, and the depiction above is a simple representation of a Squonk. ASCII art is a creative and fun way to use text characters to create images, but its complexity is limited by the constraints of the characters themselves.

.
 

J.Quondam

CR 1/8
It may have just to have any code for outputting a unicorn. Depending on the drawing libraries there may be identical or nearly identical function calls between the languages outside of the specific syntax. I don’t know TiKZ at all. But writing hello world in Java is similar to writing hello world in c++, etc.
Yeah, that's a possibility. Apparently there are snippets of code in drawing languages like TiKX for drawing animals, so I wouldn't be surprised if that's in the training data.
For the record, here are the "unicorn" images by Chat GPT as presented in the paper:

Screen-Shot-2023-07-31-at-5.02.44-PM.png


Fantasy artists have nothing to fear just yet, at least not from ChatGPT. :LOL:
 

Parmandur

Book-Friend
That is what teachers are really wrestling with. We have been teaching based on a theory of the mind that is rooted in a whole lot of cultural and metaphysical assumptions. We've known for some time that there are problems with this model, but education is a vast industry with a whole lot of inertia. Trillions of dollars and centuries worth of inertia. And this new technology is revealing that a lot of things we had assumed were uniquely special about human minds might not work at all like we thought.

We aren't close to wrapping our heads around the implications yet, or understanding what it means for education going forward. For us, this is an unparalleled existential crisis. This system with vast inertia has just hit a mighty big iceberg.
I mean, the Jesusit's worked out how this sort of parlor game works centuries ago, that'spart of the foundational work of computing theory...

Try getting ChatHPT to talk about any sort of philosophical topic, and boy do it's limitations become obvious fast.
 

Clint_L

Hero
I mean, the Jesusit's worked out how this sort of parlor game works centuries ago, that'spart of the foundational work of computing theory...

Try getting ChatHPT to talk about any sort of philosophical topic, and boy do it's limitations become obvious fast.
The Jesuits? We’ve moved just a bit past the Jesuits. Respectfully, you are missing my point. My point is that AI is easing foundational questions about both how human minds work, and how we should be training them.
 

Parmandur

Book-Friend
The Jesuits? We’ve moved just a bit past the Jesuits. Respectfully, you are missing my point. My point is that AI is easing foundational questions about both how human minds work, and how we should be training them.
I think you are missing my point, actually: these sorts of high-voltage party tricks were worked out in theory quite a while ago, and there are already theories that account for it. But that's pretty far afield from the topic st hand, of parlor trick plagiarism algorithms playing a game of Chinese Room being used to steal from artists.
 

Kurotowa

Legend
When I was a kid, I had a couple of picture books where the pages were all split into two or three sections that could be turned individually. And each book, the pictures on all the pages had a somewhat similar layout, but each in a different way. So the idea was, you could turn different sections to different pages, to mix and match them in unusual and fun ways.

That's all these generative AI programs are doing, but on a massive scale. They have a gigantic library of images stolen from the Internet, all of them cut up into tiny pieces, and a machine learning program that knows how to cross reference the metadata on the images with the text input of the users to combine the pieces in a very roughly satisfactory way. It usually gets the fine details wrong, but it's cheap and fast, and that's enough for some people.

And all that stolen art is still there in the guts of the program. It has to be. It isn't a real AI, it hasn't learned anything, it doesn't understand art theory or anatomy. All it can do is slap together quick and dirty collages from its art library, flipping those individual page sections until they line up right. So if that library was assembled by ignoring the licenses attached to the websites it scraped for them, then it's theft, pure and simple.
 
Last edited:

Golroc

Explorer
Supporter
When I was a kid, I had a couple of picture books where the pages were all split into two or three sections that could be turned individually. And each book, the pictures on all the pages had a somewhat similar layout, but each in a different way. So the idea was, you could turn different sections to different pages, to mix and match them in unusual and fun ways.

That's all these generative AI programs are doing, but on a massive scale. They have a gigantic library of images stolen from the Internet, all of them cut up into tiny pieces, and a machine learning program that knows how to cross reference the metadata on the images with the text input of the users to combine the pieces in a very roughly satisfactory way. It usually gets the fine details wrong, but it's cheap and fast, and that's enough for some people.

And all that stolen art is still there in the guts of the program. It has to be. It isn't a real AI, it hasn't learned anything, it doesn't understand art theory or anatomy. All it can do is slap together quick and dirty collages from its art library, flipping those individual page sections until they line up right. So if that library was assembled by ignoring the licenses attached to the websites it scraped for them, then it's theft, pure and simple.
I am sorry this is simply incorrect. That is not how these image generators work. They do not access metadata, they do not access the images they were trained on, they do not composite them. They are based on noise removal techniques and work from random noise. Essentially being given the task "remove noise until you have a squirrel". They do this randomly at first - and are "rewarded" by another AI (which is trained to evaluate whether an image matches the description given), until the weights reach a point where they start convincing.

It is more like a trillion monkeys putting down colors and training them to produce a specific output from a specific input.

Modern AI does not have human levels of general intelligence, but they most certainly do understand the relationship between concepts - that's how LLMs can converse in a pretty convincing manner. Something similar possible with image generation AI.

That doesn't mean it is intelligent in the sense of a human. They don't have any existence - they don't reflect or think - they process input and produce output. When they're not invoked they don't do anything. They have no memory. They can only learn by external systems changing the weights, not by agency of their own. But they are most certainly not image compositing algorithms trawling the internet and making franken-images from some database.

Here's a video explaining the techniques involved:
 

FrogReaver

As long as i get to be the frog
I am sorry this is simply incorrect. That is not how these image generators work. They do not access metadata, they do not access the images they were trained on, they do not composite them. They are based on noise removal techniques and work from random noise. Essentially being given the task "remove noise until you have a squirrel". They do this randomly at first - and are "rewarded" by another AI (which is trained to evaluate whether an image matches the description given), until the weights reach a point where they start convincing.
It’s amazing how you can be so right and so wrong at the same time.

Modern AI does not have human levels of general intelligence, but they most certainly do understand the relationship between concepts - that's how LLMs can converse in a pretty convincing manner. Something similar possible with image generation AI.
Close enough. They don’t understand any relationships. They simply have identified that relationships exist. To me that’s a big difference.
That doesn't mean it is intelligent in the sense of a human. They don't have any existence - they don't reflect or think - they process input and produce output. When they're not invoked they don't do anything. They have no memory. They can only learn by external systems changing the weights, not by agency of their own. But they are most certainly not image compositing algorithms trawling the internet and making franken-images from some database.
Yet that’s exactly where their training data came from.
 

Mirtek

Hero
The more I learn about current generative AI’s limitations and how it actually works the less afraid I am of it taking tons of knowledge related jobs. I can see it more as a glorified assistant.
That's already bad enough. AI doesn't have to replace all of the 10 accountants behind the scenes hammering away at Excel, Salesforce and SAP in their cubicles.

If it replaces 8 of them with the last two kept with the changed role of finding and correcting the few errors the AI made, that's already disruptive enough.
 

Related Articles

Remove ads

Remove ads

Top