Stable Diffusion AI Art and Twenty-Five Artists
Let’s take a deep dive into artists, given that my intentions for this week got a little derailed. It all started with Halloween. I decided to first experiment with making spooking images in Stable Diffusion (so easy with its current inability to render hands and arms well at all!) and then I went on a Halloween pumpkin kick. I had time to totally play around for a few hours and learn some interesting things, like:
- “Wet” can end up making wet reflections on the ground (streets, sidewalks) and in general they turn out really well.
- Saying “Halloween” in your prompt will get you a bevy of pumpkins.
- Stable Diffusion — as can be expected — has some strange ideas about how carved pumpkins should look.
- Having “Norman Rockwell” down as a painter will result in gothic houses to your background.
So I was pleased and posted some of the pictures to my Instagram at https://www.instagram.com/rufustheruse.art/ (along with my spooky pictures).
Then Halloween was over. I felt like I hadn’t planned ahead, waiting until Halloween itself to make spooky pictures. Okay, I don’t have any ideas for Thanksgiving, but plenty for Christmas! So let’s act like the candy aisle in the grocery store and move to Christmas.
As I played with Christmas, a theme quickly developed. I was working on Santa’s workshop elves, especially lady elves, looking scared and apprehensive. And it gave me an idea for a story so I’m grateful for that infusion of creativity. But as I worked through several versions of my prompt, I eventually came to a prompt that allowed me to experiment with a series of artist styles to see what I could learn.
Let me share some insights. I plan to show you how various artists look for the same prompt and provide links and commentary for each artist. The links should be a good home page for the artist along with a link to a Bing image search for the artist.
This took a lot of time to put together, so here’s hoping it helps someone.
First of all, here is the basis of my prompt, used in the Automatic1111 web-ui Stable Diffusion running on my local PC:
3D Vivid [handpainted:photograph:0.5] by <Artist#1> and <Artist#2> of pretty expressive santa-workshop woman warrior elf, volumetric light, ornate leather dress, fearful, dark fantasy, chaotic, intricately detailed, Symmetry, snowy, wet, Winter, Hyper-Realistic, Ultra Resolution, Dark, desolate, southern gothic, 8K, christmas, masterpiece [oil painting:hyperrealism:0.5] in the style of <Artist#3>
In this, I kept <Artist#1> and <Artist#2> constant and then would change <Artist#3> to see how powerful and impactful their style was, mixed with <Artist#1> and <Artist#2>. So, here’s an example:
3D Vivid [handpainted:photograph:0.5] by Norman Rockwell and John Singer Sargent of pretty expressive santa-workshop woman warrior elf, volumetric light, ornate leather dress, fearful, dark fantasy, chaotic, intricately detailed, Symmetry, snowy, wet, Winter, Hyper-Realistic, Ultra Resolution, Dark, desolate, southern gothic, 8K, christmas, masterpiece [oil painting:hyperrealism:0.5] in the style of Ruan Jia
So by default I’m combining Norman Rockwell and John Singer Sargent together. And looking for a rendering of a Santa’s workshop… err… warrior elf woman. It’s a thing. Above I was experimenting with Ruan Jia as the main artist.
This is based off of a prompt I’ve mentioned before as being one of my favorite constructs. You’ll note the bracketed expressions that mean to transition from starting at one place and ending at another at some percent (0.5 = 50%) of the way through the steps. In this case, it’s to start with a painting and end up with something more realistic looking. It still looks like a painting, just less stylized and more real.
Negative prompts — I want to share that I created a negative prompt for Automatic1111 over time. First of all, I didn’t want any nudie salacious images so I started with “Nipples” — that pretty much fixed that. Over time I added “frame, framed” to avoid images of framed paintings. And while I know you can’t get anything out of saying “deformed, multiple arms, extra fingers” in a negative prompt by golly it’s good catharsis to do it sometimes.
Okay, so that’s the starting point, And then, man, I partied. So many artists. So much time going through a place like Lexica and saying to myself, “Hmm, that one looks different, what artists did they use?” Then I’d go to Bing Images and do a search for “<Artist Name> art” and see what kind of results I get before trying them as the keystone <Artist#3> in my prompt. What kind of things did I learn with all these twenty-five different artists coming in as the final, <Artist#3> artist?
Alphonse Mucha — Rules!
Holy smokes, if you put in Alphonse Mucha you’re going to get super strong results. In a huge majority of the art on Lexica you see Greg Rutkowski and Alphonse Mucha put together. If you see fancy stylings with circular backdrops, you’re seeing the influence of Mucha. You can pretty much look at an image and see quite cleary if Mucha is part of the prompt. He rules a strong style and if you like it (or a touch of it) you’ll get to rock out on it.
Ayami Kojima
It was A+ crazy time with Kojima — I felt like I was getting covers to unknown Christmas stories or posters of new Hallmark holiday movies. In this case, multiple characters were showing up in the images. I even iterated on a few between adjusting the steps up and down and adjusting the “conform to what I say, AI” scale (CFG) up and down. I think my most impressive results came from Kojima for this prompt. Looking at example images I do see the occasional ensemble, so I assume that’s where this is coming from.
Donato Giancola
The next A+ crazy time output was matched by Donato — crowd scene after crowd scene was generated, which is rare from what I’ve seen in Stable Diffusion. The detail and the close-up faces were remarkable as well. I think if I was looking for a story-like setup I’d use either Donato or Kojima. Donato’s work seems focused on telling a full rich story in a small amount of space, and that really explodes outward.
JC Leyendecker
Mighty-fine leather outfits here. I like JC Leyendecker’s detail and style and think it makes for quite a few nice portraits. Typically, though, the setting is in a room or against a wall. No epic outdoor settings here. Given that Leyendecker preceded Norman Rockwell the subjects appear dated to first-half 20th century.
Charlie Bowater
You’ll see Bowater referred to quite a bit in prompts. I found I ended up with modern looking characters in interesting outfits placed within an intriguing environment. Bowater’s art is more portrait focused of fantasy subjects.
WLOP
Oh, WLOP of ArtStation and Deviant Art fame. So very popular on Lexica. Well, I can see why. The results I got are detailed outfits and beautiful faces. And interesting environments. The focus of WLOP is cute young ladies — sometimes impracticably dressed — in fantastic environments.
Frank Frazetta
Frazetta is an illustrator I have admired since I was in grade school (especially musing over the muscle and tendon structure of his painted legs). With Frazetta you’re going to get a lot of impractical outfits, muscles, and tanned skin. And sexy poses.
Bayard Wu
You can find Wu on ArtStation. I discovered Wu looking at other people’s results on Lexica and find them very interesting — super D&D content going here. Great for portraits and for getting interesting environments.
Ruan Jia
Wow. What a find. Check out Jia’s ArtStation. Jia’s great for detailed portraits with interesting lighting with either solid portrait background or a bit misty / fuzzy structure. Nice, fantastical outfits with intricate detail. I’m super happy to have discovered Jia’s art.
Jeremy Mann
You see Jeremy Mann appearing a lot in Lexica within the lists of artists. Typically there’s a subject like a person with a moody background. The subject appears modern and well rendered, though sometimes far enough back that the face is garbled. Mann’s straight art is far more filtered with a dream like quality, getting adjusted with <Artist#1> and <Artists#2> in this prompt. If you ever want a dark city street with reflective wet walkways, look to Mann.
Wenjun Lin
Wenjun Lin has a very not-too-vivid oil-painting presentation, typically with backlighting on the subject. The background is usually interesting. The subject is usually too far away to get a good sense of the face. Detail is more implied than present. Lin’s style is varied, so your results may… vary.
Jana Schirmer
Jana Schirmer has detailed, modern looking portraits that give me an early 2000s vibe. The faces are well detailed and the background has low-to-medium detail. Her art leans towards fantasy portraits.
Joshua Middleton
Joshua Middleton has a different tone than some, a more toned back richness. The details are good with conservative fashion and simple, muted color backgrounds. You can see how it’s represented in his various comic covers.
Jeff Simpson
From what I saw, Jeff Simpson made glamorous portraits with almost no background. The faces appear modern as does the dress. The occasional bit of impressive decolletage is thrown in. This really confuses me looking at some of his published art (and here). So this might be a Stable Diffusion mis-match.
Tran Nguyen
Tran Nguyen produces, in general, both interesting well-formed subjects and backgrounds. The faces and outfits lean towards modern and fantastical.
Dave McKean
If you want dreamy with the occasional bit of incomprehensible bits thrown in, turn to Dave McKean. Lots of folks including myself know McKean from various comi — err, graphic novels. Black Orchid, mmm-wha! Looking for a bit of warping to reality? McKean.
Carne Griffiths
I love Griffiths’ style. It appears so many different ways. There is usually color splashed around everywhere, though. If you want some chaotic coloring to your portraits then Griffiths is a great choice.
Artgerm / Stanley Lau
You see Artgerm referred to all the time, or sometimes Stanley Lau. For portraits, you end up with exceptional modern subjects and details. I can see why artgerm is constantly thrown into the mix of artists, especially for anything related to a superhero. Oh, anybody got $30K USD I can have to take artgerm’s year-long program?
Anato Finnstark
Anato Finnstark has come to plunge you into a vivid world of rich colors, typically with a D&D fantastical sense. Medium detailed backgrounds sometime appear, or colorful backdrops. A perfect match if you’re looking for a fantasy creature artist.
Tony Sart
If I wanted a strong D&D vibe I’d lean on Sart quickly. I’d say the outfits are a bit non-practical but matches a certain aesthetic of a little leg here, a little cleavage there.
Karol Bak
Karol Bak has a very rich, vivid, colorful presentation with modern subjects that give me 80’s vibes. If “fabulous” was part of what I was intending, I’d reach towards Bak.
Albert Lynch
Lynch produces beautiful portraits, though the subjects look dated vs. the more modern portraits you find of artists coming off of ArtStation. The luminescence of skin is amazing. If I was working on a portrait I’d try mixing in a bit of Lynch.
Gil Evgren
Hubba-hubba, a-woooooga! Sorry. Not-sorry. So Gil Evgren is basically a very talented pin-up girl artist and that comes through very strongly, between the poses and the outfits. Additionally, the subject’s appearance between hair and styling is well rooted in the 1940s and 50s. So if that’s what you’re aiming for, Lynch is a great artist to pull-in.
William-Adolphe Bouguereau
Mr. Bouguereau is going to get you some delicate ladies. Lovely portrait details but it’s to heck with the intention of your prompt. Oh you might get snow but in this case the artist is strong enough not to conform to your prompt foolishness. I do see him pop up quite often within a list of artists, perhaps to punch up the portrait’s realism.
Thomas Cole
I end with Thomas Cole, someone who impressed me decades ago with The Voyage of Life series. I would not use Cole for portraits — the images of people are going to be dated and, well, not great. But the backgrounds! The colors and details and everything in the backgrounds are amazing. If I was aiming to have a background artist I’d lean on Cole.
There are many artists I don’t list because, via Stable Diffusion where it is today, their style didn’t pop-out anything interesting. And obviously this is but a small scratch in all the artists that Stable Diffusion can create with. I’m always on the look-out for new, impressive results so please share your favorites.
