AI Generative Art

[Trike]

2 hours ago, Coyotedancer said:

 

 

So what prompts did you use to get this look? I find that trying "watercolor" and the like don't return what I'm after.

[Coyotedancer]

2 hours ago, Trike said:

 

So what prompts did you use to get this look? I find that trying "watercolor" and the like don't return what I'm after.

 

I don't remember the entire text of the prompt... It was extensive... But I'm pretty sure I asked it specifically for a watercolored drawing rather than just "watercolor".

[Trike]

Birthstone

 

 

art station style, a bald female hero comprised entirely of colored translucent crystal, bright green translucent crystal boots, bright blue translucent crystal legs with green highlights, orange and yellow crsytal knees, bright green translucent crystal torso with blue and orange highlights, green and blue translucent crystal head with slightly glowing blue eyes

 

 

art station style, action scene, a bald female hero comprised entirely of colored translucent crystal, bright green translucent crystal boots, bright blue translucent crystal legs with green highlights, orange and yellow crystal knees, bright green translucent crystal torso with blue and orange highlights, green and blue translucent crystal head with slightly glowing blue eyes, fighting a large blocky dull gray granite rock monster

 

 

art station style, action scene, city street, a bald female hero comprised entirely of colored translucent crystal, bright green translucent crystal boots, bright blue translucent crystal legs with green highlights, orange and yellow crystal knees, bright green translucent crystal torso with blue and orange highlights, green and blue translucent crystal head with slightly glowing blue eyes, fighting a large blocky dull gray granite rock monster,

 

 

 

[Trike]

The Family

 

 

The Tsoo

 

 

(Sorry, forgot to copy the prompts.)

[Trike]

Commander Quasar

 

Had a difficult time with this one. It kept adding designs to the body and didn't understand the sparkles.

 

 

art station style, no background black, cosmic hero floating straight-legged as if standing on air, wearing flat black featureless  skin-hugging tights of the deepest vantablack without any pattern,, gold with silver inlay metal knee-high boots, gold with silver inlay metal gloves and vambraces, featureless mirrored skullcap covering his eyes, stardust sparkles shimmering around his body 

 

 

[Lightslinger]

These are amazing, wow. My first run with trying to get some Lightslinger art went pretty great.

 

Is there a way to link an AI to pictures of the character then ask it to recreate in new styles/poses/etc?

 

Attached my favorites I've gotten of Lightslinger from Bing's AI.

 

 

Edited February 20 by Lightslinger

[Trike]

1 hour ago, Lightslinger said:

Is there a way to link an AI to pictures of the character then ask it to recreate in new styles/poses/etc?

Not in Bing, as I mention upthread. Might be different in other programs.

[Braddack]

 

 

my new STalker, Rad / Ninjutzu

[Lightslinger]

10 minutes ago, Braddack said:

 

 

 

my new STalker, Rad / Ninjutzu

 

Wow! Great character and design.

[sbloyd]

1 hour ago, Trike said:

Not in Bing, as I mention upthread. Might be different in other programs.

 

It appears that if you ask Copilot to generate your images, you can select one and have it re-generate them selectively.

[Trike]

3 minutes ago, sbloyd said:

 

It appears that if you ask Copilot to generate your images, you can select one and have it re-generate them selectively.

Except it always comes out different, sometimes slightly, sometimes significantly. I’d really like the option of keeping the exact image but use it in different scenes.

 

With that I could create an entire comic book that has consistent characters.

[sbloyd]

Just now, Krimson said:

It's possible with Loras though I am still trying to get better face-swapping to work.

 

Yeah, takes some legwork but it can be done, as Krimson says.  You kind of have to set up a subset of data to include with what you're working on.  You can find some of these at civit.ai, and there's examples of how to make your own on websites out there.  I've seen instructions for creating a lora file for your own likeness so that you can put yourself in a generated image (and this is how we get deepfakes, btw).

 

So you *could* create a comic book by creating a lora for each character and referencing each one for the characters present in each panel.  It's something I've been looking into but creating the source images would be a big step.  I wonder if you could use source pictures of actual COH toons, do some infill and upscale work using AI tools, and go from there; some heavy lifting in Photoshop might need to be done.

[sbloyd]

BTW, anyone know what's up with being able to select certain parts of a generated image in Copilot?  Like, I can click on and select the face, the hair, a hand, a background element, etc and it highlights it.  But I don't see what function that has, if any.

[sbloyd]

Really, if I were gonna do a comic with AI generated work, I'd still probably want to do the *actual* panels and layout myself in something like Clip Studio; you could probably even use Clip Studio's posable figures to arrange the character(s) how you want them and import them to your generator as a base image.

[sbloyd]

Chibi Blue!

[Lightslinger]

After researching a bit I found out OpenArt and Leonardo.ai both have image to image or image prompt abilities. As soon as the servers are back up I'm going to work on getting some reference shots of Lightslinger and see how closely they can recreate his in-game look.

[Techwright]

Asked Bing AI to do the impossible: think through a heap of a lot of variables to create Grill N. Chill, my fire/ice blaster.  I actually filled up all the text line a few times.  Description was constantly changing, but here's the last one: 

Jim Lee art style, dramatic pose, fire and ice auras, African superhero with short military flattop haircut, black diamond mask, orange spandex shirt with flame icon on chest, sky-blue spandex leggings, gold banded belt with black-edged trim and a row of black circles encircling the belt, black banded wristbands, small black shoulder pads, black pirate boots, black cape with sky-blue to pale-orange blend lining

 

I'm not certain why, but the A.I. immediately focused on a largely black uniform, with strong V and diamond elements, and it never let go of that.  (It probably homed in on "black superhero" which to me said ethnicity, but I guess the A.I. thought it meant spandex choices. I had to change to "African-American", and then "African" when text became tight)  It only once created shoulder pads, and it apparently has no concept of color blending, despite repeated re-wordings of that request.  Only a couple, as a result, came anywhere near what I sought.  That said, it's created some very interesting designs, and I may have to investigate to see if I can pull off one or two of the looks.

 

 

Edited February 21 by Techwright

[Trike]

1 hour ago, Techwright said:

Only a couple, as a result, came anywhere near what I sought.  That said, it's created some very interesting designs, and I may have to investigate to see if I can pull off one or two of the looks.

That one in the upper left corner looks doable, but I’m not sure that deep V pattern is available for males. But the Kid’n’Play haircut aside, those are some good looks.

[sbloyd]

1 hour ago, Techwright said:

(It probably homed in on "black superhero" which to me said ethnicity, but I guess the A.I. thought it meant spandex choices. I had to change to "African-American", and then "African" when text became tight) 

 

When the description bar becomes too short for what you need, shift over to Copilot and tell it "Create me a picture with the following prompt:" and put it there.  Copilot has a longer input limit, and will create them using the same engine.  You can hit shift-Enter after the colon to start a new line without sending the request to Copilot immediately.

[sbloyd]

Double post.

Edited February 21 by sbloyd

[Trike]

My ultimate anthropormorphic animal team-up.

 

Super Beasts + Wild Things + Sky Lords: Raptor Chapter

 

[Robothenri]

Ok, I can't resist, I had to check this out 😛

 

I signed up to the Adobe beta AI generator last year, played with it a little bit, but didn't really get far with it. So today I tried the Bing AI as a lot have mentioned it here, and I got to say, I love it! Just tried it with one character to start with, 'ALBION HAWK' DP/Martial Blaster.

 

 

Some ok-ish results to start with...

 

 

But then this came up (below) and I love it! I'm actually going to tweak the CoH costume to represent this image now. Will defo be trying this with more characters, I have millions 😄

 

[Neopatch]

It sounded like something fun to try ... so I gave it a try ...

and it wents from to

 

in Bing :

 

in Civitai :

 

I did not count the number of time the AI was blatantly not repsecting the prompt giving me things from the wrong color or forgetting a couple of them. And OMG, please who ever teach those AI, make them understand that unlesss stated otherwise, we want our characters with 2 hands and 5 fingers on each !!!

 

I'm not sure I'll continue with another toon as I did not manage to get anything good on this one 😞

 

[sbloyd]

47 minutes ago, Neopatch said:

I did not count the number of time the AI was blatantly not repsecting the prompt giving me things from the wrong color or forgetting a couple of them. And OMG, please who ever teach those AI, make them understand that unlesss stated otherwise, we want our characters with 2 hands and 5 fingers on each !!!

 

 

Yeah, the number of fingers phoenomena is a thing.  I hate when I get a perfect render, but one hand only has three fingers, or the fists have six each.  

[Coyotedancer]

Oh, the finger thing is just ridiculously annoying... I've had to crack open Photoshop and do so many hand corrections on my "Birds in the 13th Age" images... So, so many. 😖