PDA

View Full Version : Best AI for Character Portraits



dmhelp
2022-11-01, 02:55 AM
What is the best AI for character portraits? Any recommendations?

icedraikon
2022-11-01, 04:25 AM
I use artflow.ai. Throw a description into it and let it go for a min or two. Sometimes takes a few tries to get something I like and its not always great with the monstrous races.

sambojin
2022-11-01, 04:31 AM
MotionLeap on Android (and probably iOS) has a text to image AI. Sometimes you've got to be a bit more specific. The difference between "elf" and "Tolkien elf" or "DnD elf" is pretty spectacular (one giving Xmas elves, the others sometimes not). But it does know terms like "DnD artificer" for example, in classical/ manga/ whatever style, and backgrounds and stuff.

Plus you can animate them and add FX as well. Or soundtracks. On a phone. Which is handy for when you don't want it to be a big thing of actual work. Like, do it on an app, on your lunch break.

I forget how much the full version costs, but I think it's a one-off or yearly subscription of "around $50-70Aussie". Maybe it was on sale. But you do seemingly get unlimited AI images, which is nice. I do all kinds of stuff in it for my insta page, and am only just slightly leaning into the AI side of it. It does watermark them though, which is very annoying.

Maybe chuck "Portrait of a DnD *foo/character class* with *whatever* in the background in classical style" and go from there, in any AI maker thingy.


(( www.craiyon.com does it free and easy if you've got two minutes to spare.
It knows enough of a difference in the / statements, that they're totally different outcomes.
"Portrait of a DnD berserker/druid Dragonborn/halfling with mountains and clouds in the background in renaissance style" renders rather different results with similar'ish backgrounds. Faces, etc, might be janky-af, it's a pretty small AI model, but it does a thing, for free! On a webpage))

Xervous
2022-11-01, 08:15 AM
If you have a GPU that can run it I suggest getting a branch of Stable Diffusion and doing some searching for models trained on fantasy portraits.

Prompt refinement is important with any AI model.

“D&D elf” is going to be extremely hit or miss.

“D&D, portrait, male, elf, red hair, brown skin, pointy ears, handsome, double cleft chin, green eyes, in the style of pino daeni” is a fine start, to be modified with emphasis on certain keywords, additional keywords, or blacklisting other keywords (if the branch supports it) like “-beard, -pale skinned, -skinny”.

Expect to run batches of the same prompt to find a good seed to play around with.