Skip to content

AI Portrait Generation

So I've been playing around recently with using online tools to generate AI portraits. Its taken a long time to find prompts that provide a good "look" for a game like icewind dale. But when it works, you can use the same style keywords for multiple characters to generate a set of well matched, thematic potraits.

I've been using Bing Copilot Designer. And have settled on style prompts of "rpg portrait by justin sweet, soft focus, blurred, minimalist features, oil paint, drawing, in the snow, d&d, dark shading"

As examples, this process generated the following set for my long time CRPG party (which I've been using since Wizardry 8 was new)

Grimslade "Grim" (no last name): Mercenary Barbarian and Leader. Cynical Conan.
sdw6cmtakg73.jpeg

Brigit "Blur" (no last name): Speed-focused orphan with delusions of Samurai heritage
8806tvv7dpbk.jpeg

Duncan "Black" Dougal: Morbid and mildly unhinged rogue/duelist
08m4fw3jusb9.jpeg

Adam "Ace" Masters: happy-go-lucky halfling crossbow master and gadgeteer
z6rnp1rasexy.jpeg

Selina "Star" Cullen: slightly loopy hippy moon cleric
ftzg7dt61rj5.jpeg

"Sapphire": out-of-touch Elf wizard magical genius lacking in common sense
p10zcjcsmy45.jpeg

«1

Comments

  • atcDaveatcDave Member Posts: 2,240
    Those are really gorgeous! Mind if I swipe them for my own Portraits folder?
  • Zeno_42Zeno_42 Member Posts: 50
    edited June 18
    Absolutely - go right ahead. Note that they are not hard to generate yourself. That was kind of the point of the post - how easy it is to make them, An account is free. An example of the query that generated Sapphire - for example - is below:

    "dark brown skin, cameroonian elf woman, sharply pointed ears, wizard, straight dark black hair, silver facial tattoos, eyebrow rings, dark eyes, in simple purple robes with a hood, looking focused, rpg portrait by justin sweet, soft focus, blurred, minimalist features, oil paint, drawing, in the snow, d&d, dark shading"
  • atcDaveatcDave Member Posts: 2,240
    I’ve poked around with some, but nothing to show for it yet. But it’s amazing the quantity of good portraits people are generating now with AI.

    When I first started searching for new BG portraits (2001 or 2002?) there were so few to find. Remember Portrait Portal? That guy did maybe a dozen in a couple years (you still see his work, and derivatives of it all over).
    Now apparently anyone can do it. Just Google “Fantasy Portraits” and there’s hundreds, thousands even. And dozens new every week. And that’s just people who are posting what they done. It’s like we’re living with the Star Trek holodeck already.
  • Zeno_42Zeno_42 Member Posts: 50
    edited June 19
    A couple more from another party:

    Roland "Hero" DeSalle The overly-idealistic paladin:
    k2vouknnf3zm.jpeg

    Katarina "Kat" Kitovski: The cynical thief that does the dirty work behind his back:
    8jobkrnen8rb.jpeg

    Francisco "Falcon" Fernandez: The jaded mercenary duelist:
    wglw89tecib6.jpeg

    Tessandra "Tess" Tagliatova: The bard with her head stuck in the past:
    ymy3pevswmxz.jpeg

    Post edited by Zeno_42 on
  • Zeno_42Zeno_42 Member Posts: 50
    edited June 19
    A duo:

    Keira: the half-elf former street rat pickpocket trying to be something more
    i3bnwi9xzyad.jpeg

    Kord: the half-orc city guard attempting to help her do that
    h91r69f65rqt.jpeg

    Post edited by Zeno_42 on
  • Zeno_42Zeno_42 Member Posts: 50
    edited June 19
    And really old-school D&D aficionados may recognize this party - though my take on them doesn't look much like the most iconic pictures:

    awf7z4mhlfc4.jpeg

    htjwbkehohmt.jpeg

    i2brq2qv5g22.jpeg

    77sxd0wxay4j.jpeg

    2817i377lvfr.jpeg




    Post edited by Zeno_42 on
  • atcDaveatcDave Member Posts: 2,240
    Those are all terrific! I've been writing up my biographies, under the "Challenges and Playthroughs" header at left. Maybe its the more more verbal version of what you're doing.
  • Zeno_42Zeno_42 Member Posts: 50
    I'll check that out! Never noticed that sub-section before. I've always had that sort of thing mostly in my head, but since I typically cycle through the same 5-6 parties I rarely type them up any more. I know who they are. Definitely helps when making the pictures. I often fill out the image queries with personality descriptions so that I get the expressions and body language I want. Really helps make the picture feel like the character in the end.
  • Zeno_42Zeno_42 Member Posts: 50
    edited June 19
    Trying my hand at some of your characters from your descriptions in that sub:

    Telemachus
    7rwkf6t6egih.jpeg


    Prompt:
    "Mid 30s, male Human, Paladin, just over 6’ tall, muscular, brown hair, neat brown beard, and mustache. sun motif, Appealing smile, comfortable. Wearing plate armor, Pointing a sword at the sky and singing, rpg portrait by justin sweet, soft focus, blurred, minimalist features, oil paint, drawing, in the snow, d&d, dark shading"
    Post edited by Zeno_42 on
  • Zeno_42Zeno_42 Member Posts: 50
    edited June 19
    Alcina:
    2rut1mw9bjyc.jpeg

    Prompt:
    "early-30s, Human female, paladin, sun motif, 5’8”, athletic, vibrant red hair, beautiful, mature, confident, plate armor, holding a mandolin and singing, rpg portrait by justin sweet, soft focus, blurred, minimalist features, oil paint, drawing, in the snow, d&d, dark shading"
    Post edited by Zeno_42 on
  • Zeno_42Zeno_42 Member Posts: 50
    edited June 19
    Ulysses:
    9n28484adnh9.jpeg

    Prompt:
    "Human Male, early 30s, 6’8”, mercenary, muscular, wide face, broad shoulders, dark hair, full beard, moustache, scowling expression, chain armor, tapping a plain wooden staff, rpg portrait by justin sweet, soft focus, blurred, minimalist features, oil paint, drawing, in the snow, d&d, dark shading"
    Post edited by Zeno_42 on
  • Zeno_42Zeno_42 Member Posts: 50
    edited June 19
    Calliope:
    q9tbwluw3j81.jpeg


    Prompt:
    "Human female, ranger, leather armor, green cloak, Late 20s, tall, 5’9”, slim and wiry, blonde, confident, holding a lute, rpg portrait by justin sweet, soft focus, blurred, minimalist features, oil paint, drawing, in the snow, d&d, dark shading"
    Post edited by Zeno_42 on
  • Zeno_42Zeno_42 Member Posts: 50
    edited June 19
    Sybil:
    j7s7ddbckonk.jpeg

    Prompt:
    "Human female, small, 5’2", sturdy, dancer build. short brown hair, flirty expression, loose robes with low top, dancing with her arms in the air, rpg portrait by justin sweet, soft focus, blurred, minimalist features, oil paint, drawing, in the snow, d&d, dark shading"
  • Zeno_42Zeno_42 Member Posts: 50
    edited June 19
    Eurydice:
    rf3yh4nroqhy.jpeg

    Prompt:
    "female serbian human, very short, 5’ tall, long dark braided hair with decorations. wearing leather armor, Joking and laughing, holding pan pipes, rpg portrait by justin sweet, soft focus, blurred, minimalist features, oil paint, drawing, in the snow, d&d, dark shading"
  • Zeno_42Zeno_42 Member Posts: 50
    I added the prompts used to generate the images above - to illustrate how it works. You can see it doesn't always quite do what I expect. And sometimes I have to do a bit of google photo browsing to find a real-world ethnicity with features similar to what I want. And sometimes I have to iterate a bit and tweak it to get the desired results. But it gives you several attempts each go, and you can run the same prompt several times for variations until you find some you like.
  • atcDaveatcDave Member Posts: 2,240
    That’s amazing! Dude, wow. Takes my breath away!
  • atcDaveatcDave Member Posts: 2,240
    I spent some time this afternoon going through your results (and the original Telemachus story) with my wife. She has a BFA from Kendall College of Art and Design, and has some opinions about AI art...
    But she was quite blown away by your results here. She also noted, your descriptors for the AI are quite well chosen! You have a good artistic mindset (experience?) that undoubtedly helped you get these outstanding results.

    I will still give a try, but she maybe adjusted my expectations downwards...

    I also put them up on my blog here.
  • Zeno_42Zeno_42 Member Posts: 50
    edited June 19
    Heh. More engineer than artist, but I've got experience as an image analyst and I've been cajoling various forms of AI and neural networks into doing what I want since the 80s.

    For me its usually an iterative process. I start with a simple description, see what pops out, and then add additional descriptors to reinforce what I like and discourage what I don't until I get the image I was going for. Of your set, Telemachus, Alcina, and Calliope popped out quickly. Sybil took a bit more work because her description kept setting off "inappropriate content" filters, so I had to revise the description a few times to get the look without it getting the wrong idea. It kept making Ulysses look more like a brooding mage than a Little John type, so I took a few tweak cycles to get the look above. Eurydice is the most interesting.... because I think it remember previous requests in some ways each time I redo the query. I initially labelled her a "half-elf", but the features ended up looking more elven than your description. So I reverted to human and looked for a real-world ethnicity that had elvish looking features. My go to for that is below. Just picked one that looked promising, went through a couple re-submissions, and picked out the one I liked the best. But the result still had slightly pointy ears, which was a nice touch but feels like it had to have been remembering the previous queries a bit to get that.

    https://fstoppers.com/portraits/average-faces-women-around-world-2944

    The style was the hardest. I spent a long time make images of the original six trying to get the style right for a run at IWD2. Once I got a set of descriptors that consistently did a good job with that, then making images for all the characters in my library went pretty quickly.
    Post edited by Zeno_42 on
  • Zeno_42Zeno_42 Member Posts: 50
    edited June 19
    Below was my first attempt at Ulysses - where he came out looking more like Cyrano de Bergerac with a staff than your description, as I was trying get "scowling face but laughing eyes" and didn't quite emphasize his build enough.

    m943l2u3l943.jpeg
  • atcDaveatcDave Member Posts: 2,240
    Are engineers and artists so far opposite that they're a lot alike? Funny. Although actually, my father-in-law is an engineer and my wife's first job was drafting for an engineer... so she may be the one who crosses lines!

    I don't know if you looked at the link to my blog where I have the portraits I actually used, but it is interesting. Telemachus and Alcina were the two characters I played in a friend's D&D game in the late '80s, so obviously those were matched to my old memories. The other four, were partly inspired by the portraits I found. So we have a sort of circular process for them.
    Sybel was a bit of an outrageous portrait to start (when you start Googling "Fantasy Female...", you get many results that push the limits of good taste!). Still, its interesting how similar (and yet really, how much better) the portrait you submitted was to the original I'd used. I particularly like the energy and life the AI (!) gave to all of them.
    The Eurydice portrait I'd used was presumably meant to be a human rogue. But you know, 2E rules, she had to be a half-elf. And it does look like the ears on yours are slightly pointed? Which is funny in its own way. I always figured a significant number of Half-elves, Half-orcs, etc may simply choose to live their lives in human society such that "humans" may often have a range of quirks like pointed ears, unusual skin tints, stronger physiques etc. Just a highly variable populous, at least in such a high fantasy setting as the Realms. So maybe slightly pointed ears on someone otherwise human looking, might not draw any question or interest from most observers. Ultimately your effort was worth it, Eurydice is the serious stand out of the bunch of them.
  • atcDaveatcDave Member Posts: 2,240
    Oh I missed your alternate Ulysses at first. Yeah not him! But could still be good portrait art!
  • Zeno_42Zeno_42 Member Posts: 50
    edited June 20
    I used to come up with portraits the same way you describe... by endlessly searching Google, ArtHub, Pinterest, etc trying to find either images that looked like the character I wanted or just images that inspired characters by themselves. What I like about this current process - and to be clear I only started playing with this kind of AI imagery a few weeks ago - is that I can tune in exactly what I want.

    Of the images at the beginning of this thread of my own characters, some took a *lot* longer than others to get right. "Ace" the halfling took a long time to get just the right combination of Hobbit and laid back charmer. Sometimes too much Bilbo, sometimes too much Tasselhoff (or Hasselhoff). I went thru a 100 revisions of Sapphire to get just the right Asian/African/Elven mix. And getting "Kat's" cynical smirk with an edge - with a look in her eyes that says she's already figured you out - took some cycles. Or "Star's" particular combination of "not a beauty queen", "spiritual", and "just a tad too much of a wide-eyed believer" just right took a lot of iterations. "Falcon" was particularly difficult too - combining "drunken cynicism", "boredom with having run out of challenges", and "secretly wishing he still believed in something" with a Spanish falcon motif. Sort of Inigo Montoya gone sour.

    Finding existing images meeting requirements like that is close to impossible. But being able to start somewhere in the ballpark and say "now make him a little scruffier". Nah, "longer hair". How about smaller "tattoos". Etc... Allows me to dial in on a particular look in way I've never before been able to do. And once I got the hang of it I still feel like it takes me less time to come up with a portrait I'm happy with this way than it did spending forever in web searches trying to find just the right picture.

    So yeah. I would definitely encourage you to play around with it. Particularly if your wife can give you pointers on the proper Artistic terminology. I had to do a lot of poking around on google to find the right words to express art styles to make stuff that fit in with games with a strong aesthetic - like IWD or Deadfire. But as you can see, the results can really be worth it. Wish this had been around back in my own PNP days.
    Post edited by Zeno_42 on
  • Zeno_42Zeno_42 Member Posts: 50
    edited June 20
    Taking a cue from how Eurydice turned out, I decided to make another go at Grim, with a little more emphasis on letting his personality show. Add enough personality qualifiers and you can really get the AI to sell it.

    l1izry0j0891.jpeg


    Prompts:
    "muscular,male,maori,warrior,mercenary,battle-scarred,thick black hair,goatee,facial tattoos,leather armor,fierce,intimidating,calculating,menacing,standing sideways,holding an axe over his shoulder,glowering at the viewer,rpg portrait by justin sweet,soft focus,blurred,minimalist features,oil paint,drawing,in the snow,d&d,dark shading"
    Post edited by Zeno_42 on
  • Zeno_42Zeno_42 Member Posts: 50
    edited June 20
    Blur looked a little too static as well. Went for a "the only thing not in motion is her eyes focused on you" sort of look. As you can see from the prompts - describing feelings, actions, and posture really helps with the personality. And then just be willing to hit the "create" button a few times to see the full range of what it can give you with any given set of prompts.

    efnrnfo5erth.jpeg

    Prompts:
    "irish woman,late 20s,short reddish-brown hair,green eyes,athletic,scruffy,scrappy,small celtic tattoos on her face,leather armor,feisty,in motion,restless,wild,lopsided grin,jittery,slashing a short sword,ready for battle,speedy,rpg portrait by justin sweet,soft focus,blurred,minimalist features,oil paint,drawing,in the snow,d&d,dark shading,full length portrait"
  • atcDaveatcDave Member Posts: 2,240
    Yeah I love the sense of motion. Very effective.
  • atcDaveatcDave Member Posts: 2,240
    My wife found it hilarious that I’d built a musical party with no bard. So her first attempt at AI art…
    wmtxx85z9b7x.jpeg
  • Zeno_42Zeno_42 Member Posts: 50
    I like that! Looks like she's waiting at the rendezvous spot wondering where Telemachus and company went. She went south while they all went north. And that 60s Woodstock vibe is awesome! I'm still working on how to build compositions like that beyond character descriptions. How did she get AI to do things like the overlay border? I've never seen that in anything I've looked at.

    I've been working on adding personality with describing the scene better. Something like this says a lot more about "Tess" above as a character than the portrait above ever did - even though the one above captures her look pretty well.

    y1l1bom66ire.jpeg


    Also found a fly in my "AI portraits are easy" ointment. Spent hours last night trying to get a good beardless female dwarf blacksmith. That was painful. Say "dwarf" or "dwarven" and its bearded lady city. Adding "clean shaven" and the like did nothing. Go on and on about "stout" and "broad" without saying "dwarf" and you get the high school lunch lady in armor. Best I got so far was these.

    udno4dcxo5vx.jpeg

    oqhu77a427ic.jpeg



    And just try to make a Gnome that doesn't look like a Garden Gnome gone rogue...

  • atcDaveatcDave Member Posts: 2,240
    Hello, this is Dave’s wife. Thank you for your kind compliments regarding my work. I used Adobe Express to help me find an image using the words girl or young woman with a guitar, art in the style of the 1500’s, or renaissance art, and some other words that it ignored. You are correct she has a Woodstock vibe. Even though it isn’t exactly period, I liked it. I can’t remember exactly which words I used, because like you, it took many attempts and then it quit working on me. I think I also used bing. I cropped and edited the photo. I searched for a border. That also took quite a few attempts. I tried medieval, 1500s, etc. The border was edited and enhanced. I also added the text. A lot of work, but this is my area of geekiness.
  • atcDaveatcDave Member Posts: 2,240
    So my mistake, not so much AI art. But AI enhanced. But my wife has SO MANY photo editing apps on her iPad I see her scratching her head still trying to remember what all she used!
  • Zeno_42Zeno_42 Member Posts: 50
    Bing quits on you eventually if you do too much in one day. Waiting 24 hours usually resets it. And if that doesn't work going in to settings and re-selecting the language - even if its just selecting English again - seems to "unstick" it.

    I just started playing around with Artflow as well. Things I like:
    - has an explicit place to separately list things I *don't* want. Goodbye bearded lady!
    - allows you to do what it calls "train an actor". Which is essentially "remember this face/character design and then let me reuse it with tweaks to expression, position, background, etc"
    - has an explicit "make variant" function - where you tell it "start with this image I just made and then change such and such about it".

    Things I don't like:
    - I haven't yet got it to make anything resembling the soft oil painting style I've been using for IWD imagery.
    - It doesn't respond to me tweaking my prompts quite as intuitively as Bing does. With Bing even when it goes wildly off the mark I generally understand why it did what it did and how to shove it back in the right direction. With Artflow it just doesn't feel as responsive. Working with Bing is like working with an overly literal artist who doesn't speak English well. You sometimes get some bizarre interpretations, but you can usually figure out how to nudge it in the right direction. With Artflow it feels like working with a more temperamental artist that has its own opinion thank you very much.
Sign In or Register to comment.