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

[humble minion]

need the wizard from the first one in the last pic

The first wizard still has pointy ears though, and good luck getting the half-orc to be 'muscular' and still female! Yeah, I suspect you'd get something pretty usable eventually if you kept on trying, but it's more out of random chance than anything. There's clearly just holes in the logic model at this point. And that's just an artifact of the training process I suspect. You feed enough 'wizard' pics into it that are Gandalf or Fizban types, it's inevitably going to distort the model. Stereotypes reinforcing each other.

 

[SakanaSensei]

I’ve had an issue getting female characters that look older than 25 to show up. “Middle aged,” “elderly,” and ”old,” all seem to mean “slap a beard on there!” to the machine.

I’ve gotten it to work a few times, mind, it just takes corralling and do-overs.

 

[M.T. Black]

I'm enjoying all these images - thank you! I'm glad this thread has become a showcase for people's ideas. Keep them coming.

I hinted at it in my OP, but I want to acknowledge explicitly that there are some legal and ethical issues raised by these tools. I believe we (all stakeholders) must work through them and find reasonable solutions. There is no chance that these tools are going to disappear - they are simply too useful.

I'm encouraged that OpenAI, the makers of DALL·E, seem to be taking these issues seriously. In this latest version, for example, you can no longer include "in the style of [a living artist]" in your prompt (dead artists are fine). You can also no longer include living celebrities in your prompt.

Perhaps most importantly, OpenAI have promised artists the ability to exclude their work from the training data used by future models. (source)

I know these won't satisfy everyone, but they make me hopeful that we can get workable solutions to these issues.

I look forward to seeing more of your images!

 

[M.T. Black]

As a fun exercise, I wanted to re-imagine several old Dragon Magazine covers using this tool. I'll include images of the original cover plus what DALL-E generates, as well as the prompt.

Dragon #1

green dragon sitting atop a rocky hill, smoke coming from nose, high detail digital painting

Dragon #2

exterior, rocky hill with a few twisted trees, dominating scene is a barbarian with a circular shield and sword on his back, high detail digital painting

Dragon #3

futuristic space ship flies toward a planet surrounded by vast rings, dutch angle, high detail digital painting

Dragon #4

ancient contract on papyrus, sanskrit writing, margins illuminated with various images, high detail digital painting

Dragon #5

exterior, stone archway, barbarian with mighty sword battles enormous green serpent, high detail digital painting

 

[EzekielRaiden]

Okay, I have to share this one just because the AI managed to luck into some actual, honest-to-God effective image composition. I'm genuinely surprised.

Prompt was, "Dragonborn paladin in shining armor, kneeling at a forest shrine to Bahamut, painted in ink and watercolor"

Specifically, the fog and the flame. The fog gives the dark forest a spooky, otherworldly quality--but it also serves to surround the paladin with light, making him look like a light shining in the darkness. The flame, meanwhile, is almost perfectly framed by the archway that is the shrine. It subtly implies that this power, which is clearly in the paladin's hand, comes from whatever the archway is a shrine to (it's supposed to be Bahamut, but...well, the AI doesn't really understand that very well and actually put an honest-to-goodness Catholic crucifix in one of the other images!)

I was genuinely astounded at this picture. Given it's AI art, I wouldn't use it for anything serious, but I might legit actually hold onto this as a depiction of my typical dragonborn paladin characters. It's good enough to keep in one's back pocket, a few weirdnesses aside (like the weird...sword-knife-belt thing? Or the ribbons on his tail?)

 

[Parmandur]

Even Bing doesn't know what a female dwarf looks like, they are so rare.
View attachment 297113

Boy, this is a toughie, not really satisfied with any of this:

 

[Parmandur]

Elven women are easier:

beautiful elf woman Ranger knockong an arrow, myaterious forest, high qualiry digital painting

 

[Li Shenron]

Quite a shock... I played with Dall-E less than a year ago and the results were abysmal compared to these! I remember it seemed good mostly at abstract art or anything that purposefully avoided realistic human figures (e.g. when I told it "Picasso style" results were amazing, when I didn't... faces still ended up Picasso style). Must try again soon!

It would be great if it worked a bit like ChatGPT in the sense that you can ask it to modify its previous results with additional input, but I suppose that's not yet possible?

 

[M.T. Black]

It would be great if it worked a bit like ChatGPT in the sense that you can ask it to modify its previous results with additional input, but I suppose that's not yet possible?

That ability is coming. They are currently integrating DALL-E into ChatGPT and it's expected to release this month

 

[SakanaSensei]

NPC portraits are going to be so easy with this.

I’ve adjusted my earlier approach and have landed on “gender ancestry class/background, expression, black and white woodcut” and gotten the following from that very ad lib prompt.

Some silliness remains. I love the Druid’s bird man-bun.