Tuesday, July 25, 2023

Using AI Art for TTRPGs

So here's a controversal topic that will annoy the Hell out of some of my readers.

Recently I have run across several controversies over the use of AI artwork in TTRPGs. One was over a module that I reviewed where the creator used AI-generated Art on a cover and was positively brigaded for it.

Another was my recent review of the OSR game Beneath the Sunken Catacombs. I rarely go a month before I see absolute mayhem explode from some quarter of Twitter or another about someone using AI art in their product. Selling a TTRPG product that includes AI Art is guaranteed to draw backlash.

Artists who offer their creations for sale at very reasonable contracts on DriveThruRPG have pressured that marketplace so that anything using text or art generated with AI must be tagged as such.

And I aappreciate why: Art is not so much of a career as it is a calling. And it is one that is not appreciated as it should be. And one which is subject to several brutally predatory and exploitative industries. Much like writing fiction, or performing popular music. I was only beginning to gain a following for my art around 2010 when crippling migraines forced me to quit creating digital art.

Seriously, lots of love and admiration
to Dean. I wish I could fill the next
edition of Deathtrap Lite with his workd.
Honestly, I am surprised that so many talented artists like Dean Spencer offer as much of their incredible artwork as they do.

That said, I don't feel comfortable criticizing anything if I haven't tried it myself. Last month I tried installing stable diffusion on my PC, and ran into the obstacle of it being incompatible with my video card. No matter what I tried, it crashed over and over again.

After reviewing Beneath the Sunken Catacombs, however, I decided to try NightCafe for myself and learn how to use it. I quickly switched to a less restrictive tool, jumping to the Stable-Diffusion driven Imagine.

I have stuck to Stable Diffusion based technology because they ethically source their images: they do not use images without the permission of their creators unless they are in the public domain or a sufficiently open license.

This at least helps remove what was one of the early major objections to the use of AI art.

Making AI Art

A picture based on my description of my youngest
son as an astronaut using the "Dreamshaper" Model
and the "Anime v2" Style Library.
As it turns out, to use AI Art effectively, you need at least a working understanding of composition and modern art styles... there are Art skills involved. There are also literary and language skills, as effective descriptions, and tweaking them as necessary to get the AI to create what you want. Using the software to any kind of effect is a bit of a brain teaser.

One thing that is quite difficult is that within any AI Art generation is that there are a number of Models and Style Libraries that contain libraries of information on how to interpret and draw a given thing. 

Some Models are very limited,  others quite complex. The model you set when generating the image can produce vastly different results from the same written word. Often the style library supplements the Model.

Some Models and Libraries compensate for the weaknesses of the AI engine by doing cagey things like only drawing people from one angle, or making sure their hands are only visible.

An image using the exact same prompt as
above, but using the "Creative" Model and
the "Disney" Style Library.
For example, I tried  to get an image of a cartoon version of my oldest son going on an adventure with a pet squid. But none of the models available for free on Imagine know what a squid is. Nor do any of the style libraries. It is clear that at least one Model ran a search of the word "Squid" and got a rough description, but what it actually drew were Lovecraftian horrors beyond comprehension.

No combination I tried could produce aan even remotely passable squid. Although one of them would have made a fairly decent aboleth.

The process for creating an image is actually fairly time-intensive. As you need to compose an initial prompt, then generate it a number of times using different Models and Art Styles until you can find something that creates an image vaguely like the thing you want.

Then you need to change the wording of the prompt over and over again until you can find a clode match... which occasionally means swithching models and style libraries halfway throught the process to see if your adjusted wording finally clicks.

Then once you have finally landed on a relatively close image, you cycle the randomized elements a few times to see if you can't get something even closer to your goal. Finally, that will inevitably need to be touched up in another image processor.

It helps if you are using a version of Stable Diffusion or a similar engine that lets you upload an existing image (that you hopefully have the rights to) to use as a seed idea.

Sometimes you end up with dozens of images in a row before you can get even one thing you like. 

Here is an example of the evolution of a character design that nearly drove me mad:

This is Scribbles, my 3rd level Magic-user in a Lost Dungeon of Tonisborg / Blueholme Journeymanne Rules campaign I am participating in. I was chuffed that the AI usually knew what clothig a 14th century Czech yeoman  would wear... but some of the other things, like modern backgrounds, odd color splotches, mutated moustaches, etc. certainly made some options highly frustrating. and none of them really seem to know what salt-and-pepper hair is supposed to look like. And a few seem to think "sandy" means shock white. The second form the right is the one I am mst happy with, but I am going to have to correct that moustache in photoshop.

This doodle of my oldest son - based
on a photo-  took over 50
iterations to get right. Then I still had to
re-color his eyes in photoshop before
he was sarisfied.
My hat is off to anyone with the patience to make a consistent set of artistic creations using these tools.

Some things are simply beyond the technology unless you want to train it yourself, as well, and some things are heinously difficult to pull off.

For example, I wanted to create an image of an old soldier with a walrus moustache but no beard this morning. And it was a challenge beyond all reason or sanity. Some engines simply assume moustache=beard, and so would draw one, even if I said No beard, put "beard" in negative prompts, said "shaven chin", "beardless", "knobby chinned", etc. "Bare chin" set off a community standards alarm as a forbidden word that might be used to create porn.


Despite the fact that every damned picture above has an elaborate moustache and no beard.

I tried to describe a Volus from Mass
, and this is as close as I could get.
Some are even predictable. For example, if I describe a "Handsome man" the AI suddenly stops assuming clothes, aside from a pair of bluejeans. If I say "handsome man in a vest" he will only be wearing a vest. I have to say "vest and shirt" to be specific. If, on the other hand, I say "grouchy man" he will 85% of the time be wearing a hoodie.

Creating effective AI Art involves learning a lot about the quirks and assumptions the AI Itself has developed. Some of which are absolutely bizarre. For example, a "Turkish man" will always be dazzlingly handsome or grizzled I cannot get it to give me an in-between.

After spending days teaching myself the ins and outs and consulting a firend who is a wizard at this stuff, I have found my stride enough to create what I want 80% of the time within the limits of the tool. Imagine is ad-supported and will only create about 100 images for you a day. Nightcafe has no ads, but only offers you 7 free images per day before you have to start buying credit.

My son begged me to draw him Devastator using AI.
A pity, given that the source code is free.

I do know that when I create something cool and close to what I imagined, it is incredibly satisfying, and there certainly is a temptation to pay the fee to give me access to the full tool, and to do something like complete the Scurvy Knave experiment I once proposed and never had the means to create.

Using AI Art

I've essentially used my AI-generated Art for two functions: self-indulgent creations to enhance my solo gaming, and to create visual aids for my gaming group.

Solo Gaming

The first is for a science ficiton solo game that I am playing for my own edification, although I am considering using the game  for a Tale of the Manticore type podcast later.

Molly Hyaline: I haven't gotten her killed,
but it isn't for lack of my or Mythic's tyting.
In this case, my AI-Art is an imagination enhancer. It (theoretically) lets me draw chracters, creatures, and places quicly to illustrate my world. I have had particular fun drawing my PCs, with a huge volume of them dedicated to my PC Molly Hyaline, who - thanks to absolutely ridiculous luck - has repeatedly been the sole survivor of multiple disasters that have wiped the rest of the crew. The girl is like a cockroach. Seriously.

In this situation, I am doing on a tablet what I might have otherwise done with my sketchbook and a pen, to much less technically impressive results.

They really aren't for sharing except with the few friends I let in on my solo gaming experiment (and a couple of examples here). It seems like an ideal use of the technology for me.

Visual Aids

Taxu II, and alien landscape from my Solo Game
The Silver Gull Campaign has been on hiatus for two months now, mostly due to severe illness on my part, but we intend to reconvene next week. In the meantime, as I run using fuzzy 1:1 time, I am trying to handle the dozens of things my PCs have done with two months of downtime.

One place that is particularly important is that my PCs are paying for training for the crew of their ship to level them up, add skills, and grant them new weapon proficiencies. Which means my tired, and dated list of NPC henchmen needed updating. Especially as my players may be taking on some of these characters as secondary PCs, or possibly marrying one of them.

I decided that I would take a couple of hours and attempt to make a character portrait as many of the NPCs as I could. Here are some of my favourite mugshots:

In this case, my AI images are helping my players dive a little deeper into the game world. They are not being commercially published. I feel that this has the opportunity to greatly enrich my players' relationship with and sense of their henchmen and hirelings in the campaign. 

Yes, the critter in my son's hand here is
the best squid I could make Stable Diffusion draw
Neither of these are going to do any harm to the appreciation or welfare of artists. These creations have no purpose except to enhance my game, and if I did not have the ability to create them, I would be hunting my Pinterest boards for character portraits that have the sparsest resemblance to the characters I am imagining...

Or I would trawl my Pinterest and build characters based on that pre-exiting art. To no profit to the original artist whatsoever.

But What About Other Cases?

Amateur Games?

But what about my free, PWYW or $8 games on DTRPG that were created with a shoestring budget. Right now I do a small amount of the art myself, and trawl image archives like Pixabay, or choose Public Domain art to build up the content I need.

Squid are fun and easy to draw,
I did "Squid of Duality" here in 2008
C'mon SD, even I can do it.
This is fine, but the art feels wildly inconsistent, and the quality is often dubious. I am frequently disappointed by the quality of my end product, although I feel that Deathtrap Lite came out pretty well with an art budget of less than $10.

Of course, these days, Pixabay itself has become overrun with AI-generated Art, and one needs to filter it out or be painfully selective with the art they look at. It is quite possible to use AI Art completelby accident.

If I ask nothing, or next to nothing for my work, is AI Art okay? Did any of the great artists, or my competitors lose anything? I think it might be an edge case.

I do think some of the bullying I have seen of small creators using AI Art smacks of a desire to eliminte competition and seek status within one's clique. 

The stunning Gael Ghelfi was a rougish space
pirate I recruited in my Sci-Fi solo game. Finding
character art that would capture her would have
been tricky to impossible on public image archives.
Sadly, Ms. Ghelfi was taken from us too son due to
a mishap while trying to sabotage a sentry gun.
And there is another question: what if I wanted to represent characters that there is hard to get decent art of in my work? Finding decent art of gun-toting West African female space adventurers is tricky business, but I generated one for my Sci-Fi game.

Commercial RPG Products

This is the place where I am feeling things become a hell of a lot more difficult to justify.  Once you have any kind of budget, paying real artists for their work becomes a viable strategy.

From a purely utilitarian perspective, if we choose not to support the few artists who are open to working in the TTRPG niche, we will eventually make it non-viable as a market to serve. And then we will be left with no option but AI or to draw it all ourselves and see outputs grind to a halt. And some of the unique style, creativity and culture that has come up around TTRPGs as a medium.

I feel our hobby would lose something critical if the Erol Otuses, Dean Spencers, Kevin Longs, and Toni Di'Terlizzis that give our hobby life.

In the long run, I don't believe AI can replace artissts. I can make some fascinating images, but they lack the soul and energy even of my own old digital paintings.  And even if you don't believe in "Soul" one thing AI is terrible at that artists like Trampier and Di'Terlizzi did every day was innovate. AI art is terrible at making new things.

"Happy" a peice I painted in 2007

Mind you, I was able to create all those portraits in the same time frame, and without getting the migraines that forced me to stop painting for years. And I did still put work into them.

Where Are We Going?

I guess the question of "Is it Art?" is a central part of the equation here. It certainly takes skill, patience, and knowledge to create good AI art, but the conduit of brain to hand to medium just isn't there in the same way... And perhaps I am not well qualified to answer that question.

I do know that AI Art is a genie that we have already let out of the bottle. It is what we decide to consider acceptable or not - what we will buy or not - that will determine how it changes the face of the Arts and of our hobby for better or for worse. I do suspect that AI Art could make out hobby poorer in spirit if it is used in commercial products. But at the lower end, for the hobbyists... as your first steps before that big Kickstarter campaign? Im not sure. Maybe it will let the amateur create a product that will let the draw eyes they need.

Perhaps we need to set a bar? To see AI Art as a beginner's crutch? To assess it when we discuss the quality of a product? To be sure to check art credits for human names? Will these even make a difference in the end? Pay less fot AI products?

A Golden Age Comics take on Molly.
Or perhaps we need to think of it as an element of graphic design that falls into a not-art we treat differently? 

Or we learn to appreciate the skill it takes to craft an image using AI prompts and the delicate combination of Models and Design Libraries, and post-generation artistic editing as a nother form of Art; at which point we shall have to  consider it a matter of preference that we like or dislike the same way as individuals we like or dislike Modern Art? It seems more likely that this will be the way it is seen in ten years. And I might well choose to buy one book over another because I find black-and-white Trampier-style line art far more than I like slick AI design?

I also know that fear of change is a big factor in the current cultural reaction to AI Art, which will not stop its use alone. And bullying those who use it will not change their minds. Only make places like Twitter less useful. We have to decide, do we like this and see it as valuable or not in the marketplace of ideas. 

I do know, whatever the course we take, I would not be happy if we saw less Dean Spencer in the world. And fewer Erol Otuses. And for me, that is my guiding light.

Does this drive artists out of our sphere, or create more opportunities for them? Does it make it harder or easier for them to add what they add to the culture of TTRPGs? 

Perhaps that is the specific heuristic we should use when assessing the value of a given use? And perhaps, in the long term, in general.

No comments:

Post a Comment