Bringing D&D/AD&D campaign settings to life with Stable Diffusion

[Non-Edgy Gamer]

Thanks for contributing to the fall of human civilization, ASSHOLES.

Human civilization was basically dead by the late 90s. We're just burying it under AI elf drawings.

 

[Zed Duke of Banville]

Thanks for contributing to the fall of human civilization, ASSHOLES.

Saving Western civilization, one attractive woman at a time:

 

[Zed Duke of Banville]

I've belatedly downloaded the Stable Diffusion XL model, which is based on a resolution of 1024x1024 rather than 512x512 as for the earlier models. There are a variety of recommended resolutions to be used with it, e.g. 832x1216 is the closest to a 2:3 ratio that is divisible by 64 pixels in both width and height, but the higher resolutions have the drawback of a correspondingly lengthier processing time.

Challenge: 19th Century and Victorian Era characters (or whatever) more-or-less in the style of Arcanum.

I'm not specifically requesting anything, mind you.

A second attempt at creating steampunk characters reminiscent of Arcanum:

 

[Old One]

I think I like this one the best. Very close to what I envision as a Tarantian dwarf.

 

[Alex]

I think I like this one the best. Very close to what I envision as a Tarantian dwarf.

It is pretty nice, but I think the sausage fingers spoil the effect a bit.

 

[Zed Duke of Banville]

Stable Diffusion XL is much better at emulating the style of Virgil Finlay than my earlier attempt with v1.4, so here are some new portraits of the Alphatian imperial court from the Dawn of the Emperors box set:

Empress Eriadna:

Crown Prince Zandor:

Princess Mariella:

Master Terari:

Princess Asteriela of Thyatis:

Galatia Allatrian, lady-in-waiting to Princess Asteriela:

 

[Zed Duke of Banville]

When I generated images for GAZ7 The Northern Reaches, I created a few in the style of Konstantin Vasilyev and N.C. Wyeth but never followed up that style, until now:

 

[Lagole Gon]

 

[RPK]

AI has gotten a lot better at drawing hands. still not perfect, but at least it's the right number now and they generally go the right direction

 

[Zed Duke of Banville]

SDXL is also much better at the pre-Raphelite style, which I had experimented with for a few images relating to GAZ7 The Northern Reaches:

King Hord Dark-Eye of Ostland:

Queen Yrsa:

Asgrim (a cleric):

Bonus female skald:

 

[luj1]

So AI still cannot get the hands right?

 

[NecroLord]

So AI still cannot get hands right?

And eyes, it seems.

 

[Zed Duke of Banville]

So AI still cannot get the hands right?

So AI still cannot get hands right?

And eyes, it seems.

These results were generated without using even face restoration, much less other extensions, and (almost completely) without in-painting or img2img to fix specific problems. Decent enough for the intended purpose.

 

[Non-Edgy Gamer]

So AI still cannot get the hands right?

And eyes, it seems.

Eyes are easy to fix. Hands are harder, but can still be fixed with AI. Eyes take two minutes. Hands take 15 minutes to an hour, depending on how bad they are and what you want them to be doing. You might as well redraw hands in a lot of cases, and you may have to partially anyway even if you use AI to fix them.

SDXL really wasn't the leap forward in terms of quality we'd all hoped for, but it does better at interpreting most prompts.

 

[Lagole Gon]

 

[Zed Duke of Banville]

More Arcanum-esque pictures:

 

[Lagole Gon]

Drow damsels in distress.

 

[NecroLord]

Drow damsels in distress.

The ears are way overdone and exaggerated...

 

[Lagole Gon]

AI gonna AI. I've started using "half-elf" instead of "elf" to counter that, but they are still pretty big.
Well, depends on the overall direction of the prompt. They get smaller if I force more oldschool 80s art look.

Having said that, I think it looks alright on these particular pictures.

 

[Zed Duke of Banville]

AI gonna AI. I've started using "half-elf" instead of "elf" to counter that, but they are still pretty big.

Back in Stable Diffusion v1.4, I used the "small ears" prompt to counteract this trend when creating images for GAZ5 The Elves of Alfheim, but testing it this is clearly no longer sufficient. Negative prompts such as "pointy ears", "sharp ears", and "pointed ears" do help, however, more so when all three are included, and additional emphasis can be given to them.


Although the Gazetteer series ended before it could reach Sind, a country based on India and situated to the west of Darokin, the Champions of Mystara box set in 1993 did devote a handful of pages to an overview of Sind.

The capital Sayr Ulan

The Palace in Sayr Ulan

People

 

[Bruuce]

Drow damsels in distress.

Didn't I see these on /slop/?

 

[Zed Duke of Banville]

More Arcanum images, of people who might be familiar:

 

[Lagole Gon]

OC Donut Steel.

Duchess Saleosa, ruling over 4th circle of hell. Mistress of Bloodless evil, Connoisseur of Mortal Emotions.
A fallen angel praying on jealousy, unrequited passions and love tragedies.

Servant of Saleosa visits a woman in her sleep to give her free relationship advice.

Servant of Saleosa offers a youth potion. For a small price...

 

[Zed Duke of Banville]

Stable Diffusion XL is also much better at emulating the style of Edward Gorey, which is useful for Call of Cthulhu images:

 

[NecroLord]

More Arcanum images, of people who might be familiar:

Is the last one a depiction of Raven?
Or a prostitute from Madam Lil's?