This is a first in a series of guides on how to generate fantasy characters using Dall-e. This is new tech that can create useful art. It can take the ideas in your head and bring them to life! However, it is not cheap. This is a tool more like photoshop. It's FOR artists. It's not a thing to bypass them. A good understanding of art and post processing techniques will go a long way!
To the DMs who just want to make one-off character portraits:
My hope is that this guide will help you get where you want without spending too much money.
Why start with goblins?
You might ask: austinitic_steel, why start with goblins? Aren't there cooler monsters to go with?
The answer is these are creatures mostly associated with fantasy that dall-e happens to know of. Starting with humans, or dwarves or elves would be too easy because dall-e is already good at human looking things. My feeling is that goblins are going to be the most telling for most of the fantasy prompts. Also...goblins are the classic creature to see first on a fantasy adventure. It's fitting that we put them first!!
Prompt structure in this guide:
Throughout this guide we will be going over best practices for the following general prompt structure:
a goblin, [wearing], [holding], [portrait shot/medium-close shot], [artist],[flavor],[art style],[quality]
Note that the goblin is not doing anything and we haven't specified anything about hair color or other distinguishing characteristics. This is because we're trying to be simple for now. We just want to build any given goblin and see how things like artist, flavor and art style affect the prompt.
My hope is thatThis guide will inform how we treat other creatures in the dnd verse.
Let's begin :)
0. Dall-e cares about context:
Prompt: "A goblin, portrait shot"
Dall-e interprets context from your text. If there's not enough context, dall-e will give you a load of random junk for it's own reasons. Note that as of today I have no idea why it would give me two monkeys here. However, if we specify that the goblin is a man-made artistic creation, then suddenly we're in business.
1. The Art Style
First thing's first: the art style is an important part of the context of a prompt. Goblins do not typically exist as, for example, oil paintings. We will be going over the following:
1.1 Oil painting
1.2 digital art
1.3 fantasy illustration
1.4 matte painting
1.1 ---- Oil Painting
Prompt: "A goblin, portrait shot, oil painting"
Most of these (aside from the far right) look nothing like the goblins we're interested in. We need to use medium prompts that are more closely associated with fantastical creatures: "digital art, fantasy illustration, and matte painting"
1.2 ---- Digital Art
Prompt: "A goblin, portrait shot, digital art"
These ALL look like goblins to me. 100% what I Was thinking of. Dall-e is a little off with the color for digital art. Possibly because people get way more creative with digital art colors?
1.3 ---- Fantasy Illustration
Prompt: "A goblin, portrait shot, fantasy illustration"
These are pretty darn handsome goblins. 10/10 would date the far right. Not quite what we're looking for, but fantasy illustration definitely has some of the qualities we want.
1.4 ---- Matte Painting
Prompt: "A goblin, portrait shot, matte painting"
This is honestly pretty damn good too. The third one is a little too classical fantasy for my liking though. Also #2 and #4 look a little human and I'm worried this means dall-e was trained on too many matte paintings of human subjects. Adding a flavor modifier (next section) "fantasy" confirms this fear...
Prompt: "A goblin, portrait shot, fantasy, matte painting"
one of these goblins is not what it seems....
So it seems that digital art is our winner moving forward. however a close second is fantasy illustration.
2. Flavor
Next we can test some additional flavor modifiers for our prompts to get dall-e to better understand our context. we consider the following flavors here:
2.1 fantasy
2.2 high fantasy
2.3 dungeons and dragons
2.4 monster manual art
2.5 d&d
2.6 magic the gathering
2.1 ---- Fantasy
Prompt: "A goblin, portrait shot, fantasy, digital art"
This is very crisp and vibrant. I doubt the other prompts will beat this. It's probably because fantasy is just a bog standard term for lots of digital art. So here the context is strengthened.
2.2 ---- High Fantasy
Prompt: "A goblin, portrait shot, high fantasy, digital art"
well well well.....we meet again goblin I would date. These don't look right to me but they're not too bad. It might be a good one to keep in your back pocket if you'd like a wizard. Check out #2. Very wizard like imo.
2.3 ---- Dungeons & Dragons
Prompt: "A goblin, portrait shot, dungeons & dragons, digital art"
Here we get harrier goblins for some reason. It's actually quite possible that there are just as many "bugbears" as there goblins in dall-e training sets. bugbears are referred to as "hairy goblinoids". So perhaps this is still a good flavor to add but it's also maybe one that could confuse your dall-e.
2.4 ---- D&D
prompt: "a goblin, portrait shot, d&d, digital art"
The color palette here has become muted and the lines less well defined. It's likely that here there's less training data and perhaps older artwork.
2.5 ---- Monster Manual Art
Prompt: "A goblin, portrait shot, monster manual art, digital art"
Here we're again getting a muted color palette but the lines are better defined. It's possible this has better training data associated.
2.6 ---- Magic the Gathering
Prompt: "A goblin, portrait shot, magic the gathering, digital art"
These actually do look like magic card goblins to me in that there's a lot of variation. The clothing especially has something distinctly mtg in quality that I can't quite put my finger on. However, I think there's just too much variation here. Just look at the horns! Keep this flavor for when you want to get some interesting looking outfits.
To me it feels best just to stick with "fantasy" and not muck about too much with other franchises. Consider magic the gathering as a strong second for when you want to add some creative elements and consider high fantasy when you really want to hammer in the classical fantasy elements.
3. Quality:
I'm not going to talk about quality in this post. This has been well documented for photography and is an expanding study for digital art. For now know that I will be slapping on "trending on artstation, highly detailed" to all my generations.
Prompt: "A goblin, portrait shot, digital art, trending on artstation, highly detailed"
Why these phrases? mostly habit I guess....I haven't done extensive testing to ensure that these are the best phrases to use. For now, at the very least, it typically doesn't HURT us.
4. Artist:
I've done quite a lot of testing for artists in the current beta. We'll cover just a small number that I know will have some effect. Mostly the ones that dall-e knows about and which give vivid colors.
Let's try:
4.1 Stephan Martinière
4.2 Michael Whelan
4.3 Charles Vess
4.4 Marc Simonetti
4.5 Frank Frazetta
4.1 ---- Stephan Martinière
Prompt: "A goblin, portrait shot, by Stephan Martinière, fantasy, digital art, trending on artstation, highly detailed"
An artist add some distinct styling to the clothes and faces. Here he gives the goblins a devious look. They're also all eclectically clothed. Martiniere is a science fiction artist and this is going to result in issues later. Definitely a reasonable choice for some classic goblin portraits though. One strange quirk of Martiniere is that...all of these goblins are bald....Not sure why this is. Perhaps in the future everyone loses their hair?
4.2 ---- Michael Whelan
Prompt: "A goblin, portrait shot, by Michael Whelan, fantasy, digital art, trending on artstation, highly detailed"
Whelan also produces some good looking goblins. We get a goblin without clothing as though it were just a monster. The musculature is quite good here too. If you inspect the clothing closely you'll notice a lot of fantasy elements we're looking for. Whelan will actually be a clear winner later. Stay tuned.
4.3 ---- Charles Vess
Prompt: "A goblin, portrait shot, by Charles Vess, fantasy, digital art, trending on artstation, highly detailed"
Charles Vess is a comic artist and his characters are often quite stylized. This shows through here. Note the horns on THREE of the goblins. Not the right choice here. However, he'll be useful later for Tieflings.
4.4 ---- Marc Simonetti
Prompt: "A goblin, portrait shot, by Marc Simonetti, fantasy, digital art, trending on artstation, highly detailed"
Simonetti is does a lot of large landscape shots. The quality on these is going to be good but it also means that there's a lot less pulling portrait styles together. Stick to landscapes with Simonetti.
4.5 ---- Frank Frazetta
Prompt: "A goblin, portrait shot, by Frank Frazetta, fantasy, digital art, trending on artstation, highly detailed"
Frank Frazetta is a master of classical fantasy and Dall-e works very well for him. However he may not work out for some fantastical creatures. here we get some very human looking goblins. Consider Frank for your more human figures. The styles here are decidedly painterly. We'll see in later posts that Frazetta is much better suited to paintings.
5. Camera Distance (types of shots):
In this next part we'll be covering the shots, and clothing. The clothing can be very challenging with monsters since the monster is fighting against the normally human clothing and outfits. I'll share what worked and note what was challenging.
The shot here refers to the distance the metaphorical camera is from the subject. I recommend looking through the Dall-e Dictionary to learn more about all the shots you can use. This link gives a detailed overview at what the shots SHOULD mean in dall-e (that leads to the DICTION-AI-RI! definitely check it out!).
In the following sections we'll be covering the following additional shots:
5.1 close-up shot
5.2 medium shot
5.3 profile shot
5.4 over the shoulder shot
Also we've already covered portrait shots so far. After medium the picture loses a lot of detail. I don't recommend going much further past that - unless you're looking for character design inspiration.
5.1 ---- Close-up Shot
Prompt: A goblin, close-up shot, by Michael Whelan, fantasy, digital art, trending on artstation, highly detailed
Close-up shots are great for symmetry and for details. #4 may be due to how close-up shots are normally taken in photography.
5.2 ---- Medium Shot
This is where we really see Michael Whelan shine as an artist choice. We'll be comparing him to Frank Frazetta and Stephen Martiniere.
Prompt: A goblin, medium shot, by Michael Whelan, fantasy, digital art, trending on artstation, highly detailed
Whelan produces excellent goblins even at a medium shot. He'll be our clear winner moving forward. He'll also be good for fantasy outfits and backgrounds.
Prompt: A goblin, medium shot, by Stephan Martinière, fantasy, digital art, trending on artstation, highly detailed
Stephen Martiniere is a fantastic science fiction artist. While he does produce high quality art in most cases, here our goblins start to look a bit too much like aliens. Keep Martiniere in your back pocket for science fiction purposes.
5.3 ---- Profile Shot
prompt: A goblin, profile shot, by Michael Whelan, fantasy, digital art, trending on artstation, highly detailed
I can't wait to see this band. Beware that your subject may sometimes turn slightly from the camera
5.4 ---- Over the Shoulder Shot
prompt: A goblin, over the shoulder shot, by Michael Whelan, fantasy, digital art, trending on artstation, highly detailed
Excuse me but #4 is an owl not a goblin. The flaw may be due to confusion on where Dall-e needs to focus.
--------------------------------- Disclaimer:
My personal belief is that AI art should be primarily for enjoyment. Please support art and artists and furthermore write to your senators about protecting the arts through subsidies, tax breaks for working artists and for laws that protect the jobs of artists. It will not be so easy to change the pose of your characters through dall-e. Consider paying an artist for a commission of that character.