D&D General AI Art for D&D: Experiments

[Dausuul]

All I know is that when I try AI art programs, they fail and fail and fail at delivering the specific thing I ask for. Ask for some bog-standard piece of fantasy art with nothing unusual or distinctive about it, sure, they're all over that. But as soon as I start requesting particular details, they can't seem to hang onto more than one or two at a time. Get the hair right and the armor goes wrong. Fix the armor and the face starts turning into some sort of anime thing. And if I want something out of the ordinary in theme or background or anything at all, it's like suddenly I'm speaking Klingon.

AI is much better at writing code, and yet ultimately it runs into the same roadblock. It's great at giving me an example, a scaffold, a UI prototype, a test suite. It can trace through three thousand lines of code and pick out the one spot where I forgot to check for nulls. All this is super useful! But as soon as it comes time to write the heavy stuff, the working code which has to account for business needs and infrastructure considerations and resource constraints, full of very specific details that have to be right... forget it.

The AI companies have pushed the "stochastic parrot" thing to truly astonishing levels. It turns out there's a lot you can do with a sufficiently trained parrot. But a parrot it remains.

 

[Distracted DM]

All I know is that when I try AI art programs, they fail and fail and fail at delivering the specific thing I ask for. Ask for some bog-standard piece of fantasy art with nothing unusual or distinctive about it, sure, they're all over that. But as soon as I start requesting particular details, they can't seem to hang onto more than one or two at a time. Get the hair right and the armor goes wrong. Fix the armor and the face starts turning into some sort of anime thing. And if I want something out of the ordinary in theme or background or anything at all, it's like suddenly I'm speaking Klingon.

AI is much better at writing code, and yet ultimately it runs into the same roadblock. It's great at giving me an example, a scaffold, a UI prototype, a test suite. It can trace through three thousand lines of code and pick out the one spot where I forgot to check for nulls. All this is super useful! But as soon as it comes time to write the heavy stuff, the working code which has to account for business needs and infrastructure considerations and resource constraints, full of very specific details that have to be right... forget it.

The AI companies have pushed the "stochastic parrot" thing to truly astonishing levels. It turns out there's a lot you can do with a sufficiently trained parrot. But a parrot it remains.

I'd really like for this to be the case, but the linked piece Black did is pretty scary good, and it sounds like Black was able to alter details without ruining the overall picture.
Gemini is crazy good right now, it seems to be way ahead of stuff like Midjourney.

D&D General Post in thread 'AI Art for D&D: Experiments'

Tuesday at 2:03 AM

Nano Banana 3 is really a breakthrough when it comes to image control. It's the first big leap forward I've seen for a couple of years in this tech.

Here's my reinterpretation of an old Warhammer cover:

• M.T. Black

• 
• 
• 

 

[Scribe]

The cover for comparison.

 

[Bohandas]

Certainly - please share

some riffs on The King In Yellow. Most of these images are kind of old and so may be a little janky and not up to modern standards. (The fourth one in particular came out of an attempt to hide a subliminal image of Bob Dobbs in an image of a lovecraftian cult and the appearance of a character recognizable as the King in Yellow was largely coincidental)

 

[Bohandas]

Shub-Niggurath, described by Lovecraft as a "hellish cloudlike entity"

Shub-Niggurath, riffing on Quake 1's representation of as something akin to a tree made of meaty tentacles and gills

 

[M.T. Black]

All I know is that when I try AI art programs, they fail and fail and fail at delivering the specific thing I ask for. Ask for some bog-standard piece of fantasy art with nothing unusual or distinctive about it, sure, they're all over that. But as soon as I start requesting particular details, they can't seem to hang onto more than one or two at a time.

I would have agreed 100% with you up until about two weeks ago. I think Nano Banana Pro is a generational leap forward in that regard - see the image that I posted and DistractedDM re-posted, along with the prompt. The prompt is detailed and the resulting image looks good to my eye.

That's one good reason to keep this thread alive - just so that folks, especially creatives, can keep track where the technology is at.

 

[overgeeked]

I was going to avoid engaging with the discussion but it's been on my mind a lot lately. Just to get it out there, here goes.

I agree 100% with the people who say that humans create art. I just disagree with when they place that creation in the process. It's not when a person makes a thing that it becomes art. It's also rather telling that most people can't agree what is and isn't art. But we'll get back to that.

Arguing that so-called “AI” art has to be technically perfect before it could be considered art is a bad argument for two reasons. First, there will come a point when so-called “AI” art will produce technically perfect art...and it'll be sooner than we think. It's not really a counter, more like kicking the can down the street. Second, plenty of human artists have created world-renowned masterpieces that are far from technically perfect. See just about everything that's not photo-realistic painting, cubism, surrealism, etc.

Arguing that it's not art without a human consciousness intentionally creating something is also a bad argument because it throws away the long history of found art and various happy accidents that have created both wonderful art and entire movements of art. See Jackson Pollock as just one example of many.

Something becomes art when humans look at it and they consider it beautiful, inspiring, well made, when it evokes an emotional response, etc. This is also why people can't agree what is actually art. Because it's a preference. Like a fetish. You're into anime and manga, then it's art. You hate anime and manga, then it's not art. See any generational argument about music for more examples.

If you look at an image and think it's beautiful and it evokes an emotional response in you, that's art. If you then find out it's so-called “AI” art and instantly reject your previous assessment you've just given up the game.

Despite all that, I'm still 100% on the side of the people who will lose their jobs and livelihoods because of “AI” and any other form of automation. It's only in a wildly evil system such as this that these kinds of things would lead to people losing everything instead of getting days off.

 

[Jfdlsjfd]

All I know is that when I try AI art programs, they fail and fail and fail at delivering the specific thing I ask for. Ask for some bog-standard piece of fantasy art with nothing unusual or distinctive about it, sure, they're all over that. But as soon as I start requesting particular details, they can't seem to hang onto more than one or two at a time. Get the hair right and the armor goes wrong. Fix the armor and the face starts turning into some sort of anime thing. And if I want something out of the ordinary in theme or background or anything at all, it's like suddenly I'm speaking Klingon.

Which AI program have you been using? Did you try edit models to fix details?

 

[Morrus]

 

[Parmandur]

I'm seeing a bit of AI stuff on DTRPG, but it mostly doesn't seem to sell very well. The top 50 titles all seem to be hand-crafted.

But sales on DTRPG have been down pretty dramatically for the last couple of years, as many publishers will attest. Many attribute it to the new site and other changes since Roll20 took over the business.

Myself and a lot of small indies remain committed to hand-crafted commercial products - but I'm afraid I'm not seeing a huge upswell in support because of that. Quite the opposite - it's a pretty difficult time to be an indie creator. I just did a (hand-crafted) kickstarter that did half the numbers of a very similar one the year before, and my creative colleagues report similar. Lots of reasons for this (the economy probably being most important). But I was hoping the AI tide would see a premium put on handcrafted content, and that's not really happening in my section of the world.

Oh, I don't expect this to benefit indies at all: the larger companies will be the ones who market "organic art" as a boutique product. WotC ia ahead of the curve on that, I believe their anti-AI stance in products is authentic: not because they are so nice, it because they have been in the high end art prodict business for decades and know a selling point.