D&D General D&D and AI Text-to-Image Generation

[Hexmage-EN]

I typed in "Underdark" on the DALL-E demo and got some neat stuff:

 

[Guest 7034872]

That first one is great; totally reminds me of Gollum's cave.

 

[Byron.the.Bard]

Unlikely. Statistical models can become quite adept at generating patterns with which they have some familiarity. E.g. the one I linked that allows people to alter the style of an image, or request the same painting from a different angle. But there remains a key problem with such approaches, one that is unlikely to be resolved quickly or easily.

These models do not contain semantic content.

Yeah, I think that is very right and some of the failures in the post illustrate that the model is only parsing syntactic tokens. In a way, one of the challenges in using such models is a crafting a good (syntactic) string with enough tokens and hacks that the model spits out something relevant.

I would say that in many similar statistical models a critical area is where the model does not have examples and must interpolate/extrapolate. I think that is both problematic and critical: without semantic grounding the model can hardly fill the void meaningfully; but without the constraint of semantics, it can bend syntactic categories and generates wild outputs. Finding these outputs may be fun and inspiring for human artists (or for whoever like me who can not draw a beholder! )

 

[GMforPowergamers]

I immediately saw the potential for use in generating character art in D&D. It's not directly accessible by the public, sadly, but there are exciting developments on this front.

Here's a link.

thank you

edit on the wait list for the big one (can't imagine I will get in) and the mini one keeps telling me it's busy... I tried for purple hair metal winged hawkwoman just to see what it would do but no go

 

[Yaarel]

Eventually all formerly creative media will be automatically-generated by AI, probably.

Even when AI is more humanly creative, there will still be humans "directing" the creative output.

 

[overgeeked]

Byron the Bard has had fun with D&D inspired text-to-image generation!

Here's a little bit of context:



Take a look for instance at the beholders generated by an AI:

View attachment 249593View attachment 249594


To read more and see more images, check it at: D&D and Text-to-Image Generation

Something like that but for black and white line art would be amazing. How does one dive into AI-generated art? And what are the copyright issues with AI-generated art?

 

[Yaarel]

Yeah...Artflow doesn't handle non-human faces very well, does it? Like, not even tieflings, to say nothing of dragonborn.

Everything non-human looks like an elf or an orc.

It is true that every face is human, but the stylization actually looks ... artistic.

 

[Yaarel]

I wonder if there is an AI somewhere that create adventure maps on the fly!

That would be great for DMs.

 

[Beleriphon]

AI does photorealistic faces, too:

Stability AI

AI by the people for the people. We are building the foundation to activate humanity's potential.

thispersondoesnotexist.com

On that site, there are no inputs or prompts or anything; just hit reload to get a new picture. Usually you have to pore over the image pretty closely to spot the little discrepancies that show it's artificially generated. (Though once you spot some of the weirder ones, they can't be unseen, heh!)

It's a little shocking how good the tech has become-- and how fast it continues to evolve.

You know what's funny about some of those? I instantly recognized the initial input parameters. There's a particular angle that pops up for middle aged blonde women that is clearly derived from a photo of Angela Merkel.

 

[Yaarel]

You know what's funny about some of those? I instantly recognized the initial input parameters. There's a particular angle that pops up for middle aged blonde women that is clearly derived from a photo of Angela Merkel.

Regarding the Artflow.AI face generator, I suspect the "base" face is more a composite of German Americans, rather than any specific person.

A way to get around this is to name a specific celebrity − but even then, the celebrity sometimes results in a few features from the celebrity superimposing on the German-American base. It seems the German composite is for both male and female. So the resulting celebrity isnt instantly recognizable. For example, "keanu-reeves" looks less like the actor himself, and more like a German American composite.

The database only recognizes some celebrities, but not others. So it seems each datapoint was entered separately. It takes time to figure out who is in the database.

Also, if you put in a first name and a last name, it sometimes interprets this as two different celebrities, and morphs them together. It is surprising when one celebrity is a man and the other is a woman, where the result can be 1. male, 2. female, or 3. intersex.