AI Generative Art

[Trike]

4 hours ago, sbloyd said:

...and stop-motion puppet sure looks interesting!

I tried “action figure” for a couple and those were cool.

[sbloyd]

 

Infinite coloring pages via AI!

[Trike]

1 hour ago, sbloyd said:

 

Infinite coloring pages via AI!

Did you ask for colored pencils as the style?

[sbloyd]

1 minute ago, Trike said:

Did you ask for colored pencils as the style?

Black and White, Style of Lineart, and changed all the color references to white.  It turned out a lot of results that I can have fun with.

[Trike]

I tried the Runway Gen 2 that’s supposed to animate your photos and I was less than impressed. All it does is distort the image in very weird ways. Maybe in a couple years it’ll be interesting, but I don’t know how they think they can charge money for this. It’s pretty terrible, frankly.

 

Here are two of the pics I tried it on.

 

https://www.instagram.com/p/C3G2-_lteO8/?igsh=aGpmd3pranlpM3g1

[Marshal_General]

Has anyone had any luck with the Bing ai getting it to omit things? I am ryng a Half-Orc and can not get it to stop putting a nose ring in. Also telling it to do a Full Body Photo gets me at best a shot from the waist up.

[Trike]

2 hours ago, Marshal_General said:

Has anyone had any luck with the Bing ai getting it to omit things? I am ryng a Half-Orc and can not get it to stop putting a nose ring in. Also telling it to do a Full Body Photo gets me at best a shot from the waist up.

I can't get it to not add extraneous details. It also appears to remember your previous prompts, because it frequently includes things from them. 
 

The only semi-workaround I’ve come up with is to close it, wait a few minutes, then ask for something completely unrelated. Then I go back to my original ask. That works maybe a quarter of the time.

 

For example, I asked for a superhero in the style of Vitruvian Man, which it never really did, so then I tried creating my Jumping Jack Slash character who is just a pumpkin head with arms. And for some reason it started giving me four- and six-armed versions… just like Vitruvian Man. So I closed it out, came back and asked for a hippy hero, then asked for Jack again and it gave me a regular biped.

 

Weird.

 

Jack:

 

AI versions:

 

Hippy reset:

 

Back to Jack:

[sbloyd]

2 hours ago, Marshal_General said:

Has anyone had any luck with the Bing ai getting it to omit things? I am ryng a Half-Orc and can not get it to stop putting a nose ring in. Also telling it to do a Full Body Photo gets me at best a shot from the waist up.

 

I include NO CAPE in my prompts but it randomly puts them in from time to time anyway.  Some generators have a "negative prompt" - things you want omitted - but the Bing generator doesn't have this feature.  Some generators also have a CFG weight that determines how strongly the generator adheres to the prompt; most defaults give it some 'leeway'.  Alas, Bing doesn't have this feature available to us either.  The user end is pretty stripped-down... my main interest in it is that it's not blocked at work for me 😄 😄

[Techwright]

Next up my mace/fire brute, baseball-themed "Heat Hitter".  I couldn't get the A.I. to understand I wanted wrap sunglasses and a 1920s style uniform, but I did like the creative results.

 

Incidentally, the top corner to viewer's right, I instructed the A.I. starting "Michael Bay explosions style..."  😁

 

[Trike]

6 minutes ago, Techwright said:

instructed the A.I. starting "Michael Bay explosions style

I like how it has the explosion behind him yet also reflected it in the glasses. 100% Michael Baysplosion. 😂

[Techwright]

One other of A.I. Heat Hitter.  The rest of it is okay, but I included it because the shirt logo is so cool.  It's the epitome of "hellcat" and I'd love to see it added to game costume options.

 

[Trike]

Tried creating some Norman Rockwell-inspired pics with superheroes, and the ones in the soda shoppe were delightfully wholesome. As soon as I placed them in a playground things took a turn.

 

This is the prompt: Norman Rockwell style, two little kids dressed up as pulp superheroes in a playground

 

I'm captioning this first one, "I shot a man at recess just to watch him die."

 

 

"So I told the dame, 'You mouth off again and you get the back of my hand.' She sent me to bed without supper."

 

 

Spengler & Robin

 

Edited February 10 by Trike

[Marshal_General]

If the character creator allowed me to make this, I would.

 

[Trike]

2 hours ago, Marshal_General said:

If the character creator allowed me to make this, I would.

 

 

 

 

What is his and/or her name?

 

In the meantime, please meet Dread Panda Roberts. (Eyepatch optional.)

 

 

[Marshal_General]

34 minutes ago, Trike said:

What is his and/or her name?

 

In the meantime, please meet Dread Panda Roberts. (Eyepatch optional.)

 

 

No name yet. I had an idea for a fairy that turns into a were red panda. Since I am a purist, the fairy would be female.

[ThaOGDreamWeaver]

Playground AI can also use AI to expand on your prompts and come up with interesting stuff.

 

Trying out some takes on DW with Playground v2's Fantasy Worlds preset:

 

 

vs StableDiffusion XL with Masterpiece:

 

Out of the two, #1's much closer to what I wanted, minus the hangover redeye, though #2 is probably going on my ParagonBumble profile.

(and hey devs... I still want my jet bike flying mount when you get some time x x)

Edited February 11 by ThaOGDreamWeaver

[Techwright]

This one has been a tough wrangle:  Vincent Poe, my young dark/dark corruptor.   Early attempts produced dramatically wrong results.  Even now, I'd say the system only understands about 80% of what I'm after, but at least it's keyed in thematically now.  I've yet to get it to understand the concept of the white-faced goth look without it immediately adding a boatload of undesired extras to the look, and it seems to have no idea how to render a brass collar, so I had to leave those off, but these latter works were an improvement.

[Trike]

8 hours ago, ThaOGDreamWeaver said:

Trying out some takes on DW with Playground v2's Fantasy Worlds preset

 

I like how she’s flying between worlds - 1940s western city on the left, older middle eastern city on the right.

 

That second pic though… oof. She’s sitting on the gas tank and those handlebars are the very definition of random shape. It’s so weird how it does things like this. I’ve done some pics the past couple days which are interesting at first glance but upon closer inspection the characters have extra hands or melted faces or random things sticking out of people.

[Trike]

5 hours ago, Techwright said:

This one has been a tough wrangle:  Vincent Poe, my young dark/dark corruptor.   Early attempts produced dramatically wrong results.  Even now, I'd say the system only understands about 80% of what I'm after, but at least it's keyed in thematically now.  I've yet to get it to understand the concept of the white-faced goth look without it immediately adding a boatload of undesired extras to the look, and it seems to have no idea how to render a brass collar, so I had to leave those off, but these latter works were an improvement

It’s so weird, but we’re in early days of this stuff. Sometimes it gets it perfect the first time out, sometimes you have to do 1,000 iterations just to get something that approximates what you’re asking for. Last night I never could get it to understand what I meant by an exoskeleton. Just kept giving me metal bones, no matter what I tried.

 

One thing that would improve the usefulness of these algorithms would be the ability to select an element from a generated picture to replicate in subsequent images. I don’t know how often I get a character right but the scene is wrong, or that there’s a random item in the background that’s cool but nothing else is suitable. If we could just pick out that one thing, whether it’s a character or an object, and use it going forward, it would change these programs from a toy into a tool.

[Trike]

_____________________

TEAM FACEPUNCH

********************

 

The Big Galoot, Cliff Hanger, Danger-Man, Nuckle Sammich, Throwndown

 

 

 

 

 

 

 

 

 

 

 

 

[Techwright]

On 2/11/2024 at 2:23 PM, Trike said:

One thing that would improve the usefulness of these algorithms would be the ability to select an element from a generated picture to replicate in subsequent images.

 

So did you find something that accomplished that?  Your team shots seem to imply "yes".

[Trike]

19 minutes ago, Techwright said:

 

So did you find something that accomplished that?  Your team shots seem to imply "yes".

 

Sadly no. At least not in Bing AI, which is what I've been using.

 

I just did hundreds of iterations until I happened upon a few which kinda-sorta match up. I found that having more than two characters in the scene *really* confused it, so it would mix and match descriptions.  More than a third of the time it would also add other characters for some reason. So I couldn't get it to render a team shot.

 

To get the final team shot I kept asking for solo pictures until I had some that were close enough to the game look, then I erased the extraneous bits and edited them together.

 

No two versions of Cliff Hanger look alike, but I figure the cowboy outfit disguises it a bit. (The program has no idea what chaps are.) Throwdown came out with a full head of hair 90% of the time despite specifying that he was bald. On the other hand, Big Galoot came out bald most of the time, and I only got him wearing blue sunglasses once. (Which I used above. He was also shirtless, so I just color-swapped some of his skin for the orange shirt as a dumb kludge. Don't look too closely, it's terrible.) I NEVER got a blonde version of Danger-Man, no matter how many times I asked. Not even with solo shots.

 

And whenever I had them fighting bad guys, half the time the algorithm would turn them into one of the enemies. That shot with Danger-man and Nuckle Sammich facing the werewolf came from the prompt of those two and Galoot kicking werewolves off a moving train. Instead it gave Danger-Man a beard and turned Galoot into a wolf. In the shot with Throwdown and Danger-Man fighting plant monsters, it gave Dan a domino mask for some reason and also gave him Throwdown's utility harness and belt. With no plant monsters! But it looks cool, so why not.

 

Here are a few of the many (many!) missed ones:

 

[Oz The Elderly Gaymer]

So these prompts you've been adding have been so helpful! I really couldn't figure out the 'right words' until I saw your examples.  So, here's some for my Praetorian Dual-wielding Blaster, Markus Magnum:
 

Quote

cel shaded, style of artstation, background of a gold 5 pointed star in soviet art style, mercenary, trimmed beard, black button-up shirt with sleeves rolled up, grey armor plated leather vest, black hair, wearing sunglasses with 1950s style black frames and cobalt blue lenses, grey khaki pants with  blue stripe, black work boots, full body shot, action shot, martial arts, sharp focus, huge male, muscular rugby player build, dual beretta 92fs heavy pistols

 

Ironically, when Bing started adding suspenders instead of the armor-plated vest, I stared at the results, fanboi-shrieked 'OMG!' and ran off to replicate them.  Unfortunately... no suspenders.  Boo.

Edited February 15 by Oz The Elderly Gaymer

[sbloyd]

1 hour ago, Oz The Elderly Gaymer said:

So these prompts you've been adding have been so helpful! I really couldn't figure out the 'right words' until I saw your examples.  So, here's some for my Praetorian Dual-wielding Blaster, Markus Magnum:
 

 

Ironically, when Bing started adding suspenders instead of the armor-plated vest, I stared at the results, fanboi-shrieked 'OMG!' and ran off to replicate them.  Unfortunately... no suspenders.  Boo.

 

There's a chest harness with ammo pouches that's kind of suspenders-like.

[Oz The Elderly Gaymer]

13 minutes ago, sbloyd said:

 

There's a chest harness with ammo pouches that's kind of suspenders-like.

 

Oh I know. The pouches are just too distracting to me, and I never found anything similar.