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

[orcinator]

Wonder how well it would handle making cyberpunk/scifi NPCs. I already see those pieces of art plaguing artstation and pixiv though I wonder how good it is at drawing specific prompts.

 

[Zed Duke of Banville]

Luna the Silver Heaven, first layer of Mount Celestia, an ocean with mountain-islands containing citadels of polished white stone:

Bahamut's Palace in Mercuria the Golden Heaven:

Venya the Pearly Heaven, where the slopes are terraced fields:

Solania the Electrum Heaven, with icebound peaks:

The First Monastery of the Planes-Militant:

Empyrea the City of Tempered Souls, in Mertion the Platinum Heaven:

Archons:

 

[Zed Duke of Banville]

The first layer of Baator is Avernus, a battleground with blasted, rock-strewn fields under a crimson sky:

The second layer is Dis, ringed by mountains but almost entirely occupied by the City of Pain, comprised of blackened iron walls smoking with intense heat:

Minauros, the third layer, is a stinking bog, with polluted rain falling from the leaded sky:

Phlegethos, the fourth layer, is filled with rivers of fire, magma lakes, and lava flows:

Stygia, the fifth layer, is a frozen sea with crushing ice floes and mammoth icebergs, and constant lightning strikes:

The sixth (Malbolge), eighth (Cania), and ninth (Nessus) layers are less amenable to visual representation.

Maladomini, the seventh layer, consists of ruined cities, slag heaps, and strip mines belching filth into the darkened sky:

Erinyes are a lesser devil baatezu, first appearing in the 1977 Monster Manual:

Edit: I should have included as a comparison Tony DiTerlizzi's illustration of an erinyes for the first Planescape Monstrous Compendium Appendix:

 

[NecroLord]

I did not realize the Erinyes had such BIG BAZONGAS.

 

[Dexter]

If you're using the WebUI, have you played around with "Highres. fix"? It doesn't always bring good results, but it could work fine enough for landscape stuff, and when it does it usually gets rid of that pixelated/hazy low-res look. Even a slight resolution increase like 896x640 seems to help, or maybe something like 1280x1024 if you have enough VRAM. Don't forget to Set an Upscaler for img2img in Settings like Remacri before.

 

[Zed Duke of Banville]

Wonder how well it would handle making cyberpunk/scifi NPCs. I already see those pieces of art plaguing artstation and pixiv though I wonder how good it is at drawing specific prompts.

Works for cyberpunk:

 

[NecroLord]

Nice boobs.

 

[Lagole Gon]

>Overly sharp.
>Smooth airbrush-y light gradients.
Please... stop.

Sorry fleshbags, this is how a REAL cyberwoman looks like.

 

[Zed Duke of Banville]

Arborea's first layer includes Arvandor, the High Forest, land of the elves, a realm of ancient trees and cathedral-like groves:

The first layer is better known as Olympus, named for a realm of lush hills, fertile vineyards, olive groves, and orchards, as well as whitewashed towns such as Thrassos, on the coast of an azure sea:

The second layer, Ossa, is a vast yet shallow sea, scattered with islands, though it also contains some deeper underwater realms such as Caletto, home of Poseidon:

Pelion, the third layer of Arborea, is a dusty place of white sand, containing little other than the dun-colored stone palace of Nephythys:

Eladrin are the Arborean equivalent of Archons:

 

[Lagole Gon]

Landscapes are okay, but the characters look super gay and like semi-animu artgem shit.

 

[Zed Duke of Banville]

Landscapes are okay, but the characters look super gay and like semi-animu artgem shit.

Yes, these Planescape characters have been created with the usual artist combination. I'm being more selective in my choice of artists for the Ravenloft setting, but first a final look at Planescape, namely a few of the infinite layers of the Abyss:


The Demonweb Pits of Lolth:

Smaragd is a realm of emerald jungles, acid rain, and fermenting poisons:

The Rotting Plain is a pungent expanse of baking dry savanna and shallow swamplands:

The Slime Pits, formally called Shedaklah, are a bubbling morass of fetid sludge, caustic slime, molds and fungi:

Thanatos is a frozen necropolis under a moonlit sky:

Torremor is a seemingly endless verticality of beams, perches, pillars, ramps, and stairs:

And the Abyss wouldn't be complete without Succubi (one of whom will eventually change alignment almost to the opposite of chaotic evil):

Tony DiTerlizzi's succubus from the Planescape Monstrous Compendium Appendix I:

 

[NecroLord]

That's how succubi should look like.
HUGE FUCCIN BOOBS!

 

[Zed Duke of Banville]

For the Ravenloft campaign setting of gothic horror, I directed Stable Diffusion to an art style based on Henry Fuseli, Zdzislaw Beksiński, Nicola Samori, and H.R. Giger.


Azalin the Lich, lord of Darkon:

Adam, a monster created by Frankenstein Mordenstein:

Strahd von Zarovich, lord of Barovia and antagonist of the original Ravenloft adventure module:

Vlad Drakov, lord of Falkovnia, former leader of a mercenary band called the Talon of the Hawk:

Bastellus, a haunting undead creature that comes in the night to feed on the dreams of helpless sleepers:

Blood elemental, with a roughly humanoid appearance and tentacles of blood:

Mist Horror:

Gargoyle:

 

[NecroLord]

For the Ravenloft campaign setting of gothic horror, I directed Stable Diffusion to an art style based on Henry Fuseli, Zdzislaw Beksiński, Nicola Samori, and H.R. Giger.


Azalin the Lich, lord of Darkon:

Adam, a monster created by Frankenstein Mordenstein:

Strahd von Zarovich, lord of Barovia and antagonist of the original Ravenloft adventure module:

Vlad Drakov, lord of Falkovnia, former leader of a mercenary band called the Talon of the Hawk:

Bastellus, a haunting undead creature that comes in the night to feed on the dreams of helpless sleepers:

Blood elemental, with a roughly humanoid appearance and tentacles of blood:

Mist Horror:

Gargoyle:

All of those look pretty impressive.
It seems the Stable Diffusion is better at creating Undead instead of cross-eyed busty elven and Tanar'ri bitches.

 

[udm]

Zed Duke of Banville I managed to get Stable Diffusion running using this tutorial:



But I'm getting terrible results. For example, typing "elf warrior from mystara" gets me some North Pole-looking elf. No idea what I'm doing wrong, thoughts?

 

[rusty_shackleford]

easiest way to use it is to just use colab
https://colab.research.google.com/drive/1kw3egmSn-KgWsikYvOMjJkVDsPLjEMzl

just requires you to press the arrow button + enter your API key from huggingface

 

[Non-Edgy Gamer]

Zed Duke of Banville I managed to get Stable Diffusion running using this tutorial:

I don't know why github nerds insist on complicating this crap for normies when you can just go to Code and save a zip file, and it will work just fine.

Sure, it might make rolling back updates or whatever harder, but most people will probably never even update it anyway, let alone checkout an older version etc.

 

[Zed Duke of Banville]

But I'm getting terrible results. For example, typing "elf warrior from mystara" gets me some North Pole-looking elf. No idea what I'm doing wrong, thoughts?

If you aren't referring to any specific artists or style in the prompts, or even to something such as artstation that might provide decent artwork for a baseline, then you will probably have terrible results. For elves, you might consider ending with the prompts "D&D, fantasy, artstation, Hearthstone, art by Artgerm and Greg Rutkowski and Alphonse Mucha". Also specify "small ears", since the training data must have included a large number of images with those horrible Japanese-comics-style elf ears.

 

[Lagole Gon]

Monsters.

 

[udm]

If you aren't referring to any specific artists or style in the prompts, or even to something such as artstation that might provide decent artwork for a baseline, then you will probably have terrible results. For elves, you might consider ending with the prompts "D&D, fantasy, artstation, Hearthstone, art by Artgerm and Greg Rutkowski and Alphonse Mucha". Also specify "small ears", since the training data must have included a large number of images with those horrible Japanese-comics-style elf ears.

Thank you kindly, I'll try that later!

I don't know why github nerds insist on complicating this crap for normies when you can just go to Code and save a zip file, and it will work just fine.

Sure, it might make rolling back updates or whatever harder, but most people will probably never even update it anyway, let alone checkout an older version etc.

I haven't got the damndest idea how github works lol. I'm just glad I finally got Stable Diffusion working. But yeah would have been nice to go through fewer hoops.

 

[Dexter]

I haven't got the damndest idea how github works lol. I'm just glad I finally got Stable Diffusion working. But yeah would have been nice to go through fewer hoops.

That video seems retarded tbh. For instance you need neither a GitHub Account, nor a "Hugging Face" account. For the models you can just torrent them from here: https://rentry.org/sdmodels
Installing Git makes sense though, since Updating is as easy as opening the console and typing "git pull" and there's usually several Updates and even new Features every day at this stage.
The description of how you're supposed to do that is also pretty retarded and overly-complicated though with creating folders and moving shit around and whatnot, all you need to do right-click on the drive you want to install it and choose "Git GUI here", then click "Clone Existing Repository" and enter the GitHub URL and a folder name and you're done.

 

[Zed Duke of Banville]

More Ravenloft:


Quickwood, a living tree that eats people:

Reaver, an underwater humanoid monster:

Illithid vampire:

Meredoth, a necromancer, dark lord, and antagonist in adventure module RA2 Ship of Horror:

Senmet, a greater mummy appearing in adventure module RA3 Touch of Death:

"Aggie", an undead sea serpent appearing in the Castles Forlorn box set:

 

[SleepyBlue]

Was there a guide you followed to get a local version of Stable Diffusion running?

You can find various guides online, but the basic steps:

1. Download Stable Diffusion and extract to your drive --- note the README file contains useful information
2. Download the latest Checkpoint data for Stable Diffusion (requires an email address, even if a fake one) and save it as [drive\mainfolder\]stable-diffusion-main\models\ldm\stable-diffusion-v1\model.ckpt
3. Download and install Miniconda (I first tried with Anaconda, but this caused problems)
4. Download and install the latest version of Python, if you don't already have it
5. Open a command prompt in Miniconda, change directory to [drive\mainfolder\]stable-diffusion-main, and then enter the statement conda env create -f environment.yaml (this should be necessary just once)
6. At this point, any time you want to run Stable Diffusion, you can just
   1. Open a Miniconda command prompt
   2. Change directory to [drive\mainfolder\]stable-diffusion-main
   3. Enter the statement conda activate ldm
   4. Start running your art generation commands, such as python scripts/txt2img.py --prompt "Duke Stefan Karameikos, monarch, royalty, fiery red hair flecked with gray, fiery red beard flecked with gray, sea-blue eyes, gruff, candid, uncompromising, William-Adolphe Bouguereau, Edmund Leighton, Frank Dicksee, John William Waterhouse" --plms --n_iter 5 --n_samples 1
   5. Note you can change the seed number by adding --seed #, where # is whatever number you want to enter
   6. Output appears in [drive\mainfolder\]stable-diffusion-main\outputs\txt2img-samples\samples; also, a single image containing all five images created from a seed appears in the folder above that

For GAZ5 The Elves of Alfheim, I continued using the Mucha/Rutkowski/Artgerm combination; although it should be possible to generate elves and dwarves in a specific artistic style, it would be more difficult. There is a section on notable personalities, which includes notes on appearance and personality, and I also requested a forest background in the prompts:


King Doriath Erendyl:

Brendian Erendyl:

Lynnwyl Chossum:

Dyradyl Feadiel:

Arloen Treelife of the Grunalf Clan:

Durifern Widefarer of the Grunalf Clan:

Beasthunter of the Longrunner Clan:

Coolhands of the Longrunner Clan (my favorite of the GAZ5 portraits)

Mealidan Mealidil:

Redsword Truetalker of the Red Arrow Clan:

Brightsword of the Red Arrow Clan:

Shurengyla is a stereotypical blonde elf-maiden, so a few of her better portraits could be used to represent different elf-women:

Hey there, I really want to generate my own drow elf, but its like impossible, after hours of trying I dont get nothing decent. Could you share some of the prompts that you used? This images look freaking awesome!

 

[Zed Duke of Banville]

Hey there, I really want to generate my own drow elf, but its like impossible, after hours of trying I dont get nothing decent. Could you share some of the prompts that you used? This images look freaking awesome!

I hadn't tried to generated a drow / dark elf, so I can't share prompts appropriate for them rather than for regular elves. As mentioned above, for the elves I ended my prompts with D&D, fantasy, artstation, Hearthstone, art by Artgerm and Greg Rutkowski and Alphonse Mucha, and I also included an earlier prompt for small ears since there seems to be training data with those ridiculously-oversized Japanese-originated elf ears.

Stable Diffusion might have difficulty generating drow with the proper combination of skin color and facial features. Taking the prompts for Coolhands and then replacing "elven woman" with "drow woman", "elf" with "dark elf", and changing the skin and hair color to dark and white, respectively, yielded this as the best of 30 images, many of which had skin color that was normal for elves rather than drow:

 

[Rincewind]

This is philosophically interesting, because most (if not all) of creativity is just recombining things that you've already seen and experienced, plus some elements of either consciously controlled or accidental randomness. I can certainly tell you that when doing art, many of my most creative ideas are coming from random experiments, or when I'm in an "altered state of consciousness" (don't necessarily think of drugs; it can come from some atypical physical state, such as illness, or dreams, which are more or less some noise generated by your neurons while asleep).

People who are considered "creative geniuses" are just really good at combining disparate things in novel and striking ways.

I think what these AI experiments will show us in the long run is that certain aspects of "humanity" are overrated and over-mystified; in reality, there are just a few simple processes at play with some randomness involved.