Your thoughts on A.I. art creation

Started by Racoon, Sun 07/08/2022 21:08:14

Previous topic - Next topic

Mandle


LimpingFish

Quote from: Snarky on Fri 12/08/2022 18:08:07
* With one exception: ArtBreeder. That offers enough knobs and dials to effectively let you tweak the results towards what you're looking for. Within the scope of what it does, e.g. character portraits, I think it's already a viable alternative to hiring an artist (or at least to do the bulk of the work before a final polish by a human).

That's a fascinating tool for sure, but I'm not sure how I feel about it in practice. As a way to generate concept art, or rough ideas, it might come in handy. I'm not sure if I'd be willing to pay for it for that purpose, though.

I certainly doubt I'd use it, or any AI-generated work, to create a final piece of art that I would then claim as my own. I may be naive, but it just doesn't feel...honest. :-\
Steam: LimpingFish
PSN: LFishRoller
XB: TheActualLimpingFish
Spotify: LimpingFish

Mandle

I guess if any artist comes under that kind of "Milli Vanilli" scrutiny then people could just challenge them to create a new work in real time with people watching the whole process.

Snarky


cat

Very interesting read, Snarky, thanks for sharing!

Gilbert

Too lazyDidn't have time to read the whole thread and I don't know whether it's suggested or even done already by anyone, but I think without worrying morally ambiguous use of the technology in aggressive ways (such as generate the whole story) there can still be a lot of fun use of it.

One idea is to use AI to generate pictures in real time during a playthrough of a pure text adventure (e.g. the good o' Zork series).
This way, you are still playing the same old games but would have some random eye candies to look at.
It'd be nice if someone tries to make an IF parser with such functionality.

LimpingFish

Somewhat depressing article over on Kotaku.
Steam: LimpingFish
PSN: LFishRoller
XB: TheActualLimpingFish
Spotify: LimpingFish

Snarky

The Algorithm has figured out that I'm interested in this topic, so now it keeps throwing new articles about it at me every day. Here's one that might interest you: a guy used the AI tool Stable Diffusion to "upgrade" pixel art into photorealistic images:



(If I understand the process correctly, it's actually more like generating a new image from a detailed description, using the pixel art as a guide.)

I think you'll agree that the output is pretty goofy-looking (King Graham cracks me up!), but it's also pretty convincing. I probably wouldn't have spotted most of these as fake (or at least, not as AI-fakes) without close inspection.

Meanwhile, another guy entered a digital painting generated by Midjourney (and then upscaled with Gigapixel A.I.) into an art contest at a state fair, and won first prize in his category:


eri0o

Midjourney must have been trained with some Korean/Chinese artists. I used to follow some that did lots of art for RPG and rereadings of MtG cards, and I remember they used to look a lot like this picture.

Mandle

Interesting that an A.I. can produce a genuinely awesome Lovecraftian entity considering that they are always described as something beyond the comprehension of the human mind in the stories.



Mouth for war

#31
I just "made" these with Nightcafe. I think they look pretty cool :D













And this from StarryAI



Not THAT far from being usable IMO
mass genocide is the most exhausting activity one can engage in, next to soccer

CaptainD

Quote from: Mouth for war on Fri 09/09/2022 20:08:44
I just "made" these with Nightcafe. I think they look pretty cool :D

Indeed, they look pretty good.

Quote from: Mouth for war on Fri 09/09/2022 20:08:44
Not THAT far from being usable IMO

I think the difficulty would be not just getting a scene but also getting the viewpoint / perspective that you want, although this could possible be achieved to some extent by judicious cropping.
 

Mouth for war

"I think the difficulty would be not just getting a scene but also getting the viewpoint / perspective that you want, although this could possible be achieved to some extent by judicious cropping."

Yeah you're absolutely right :-)
mass genocide is the most exhausting activity one can engage in, next to soccer


Danvzare

Quote from: TheFrighter on Mon 24/10/2022 08:47:54An example of game based on an AI-generated story:

https://meiri.itch.io/bohemian-symphony

_
Do they ever say what they used to generate the AI-generated story?

After all, saying something is made by AI is one thing, proving it is another. And a lot of people seem to be using it as a way to jump onto the hype-train and grab some attention.

I'm sure it is indeed based on an AI-generated story, but considering I've seen people repost pictures from a decade ago, stating that it was made by AI, when in fact I know that it wasn't, has made me skeptical with these things.
(Even more so when they say they used a certain tool, and whenever you use that same tool your results are barely decipherable messes, while there's somehow looks perfect.)

KyriakosCH

I am not a fan of winning art contests (meant for humans) with an AI-image...  (nod)
Those images do look nice, though!
This is the Way - A dark allegory. My Twitter!  My Youtube!

Retro Wolf

#37
The first two images are basic scenes made by me in Blender, fed those into Stable Diffusion.







These two processed with less "artistic freedom".








Very cool technology!

Snarky

A discussion in another thread made me ask myself if an AI bot could convincingly compete in the Background Blitz on its own, and decided to test it. Spoiler: Probably not at present, without some human help.

First, here's the result, on the topic "spooky":



I think that looks decent enough. But I found that I had to perform human intervention at several steps along the way to achieve this.

Spoiler
I first thought I would just feed the Blitz topic as a prompt to one of the image generating neural nets, like Dall-E or Midjourney. But since the topic this month is just "a spooky background," I decided I needed something more detailed, so I used GPT-3 to generate a short text from the prompt:

Write a description of a spooky painting from a point-and-click adventure game.

This was its initial response:

Quote from: GPT-3This painting is of a dark and spooky forest, with black trees and a red moon. There is a small path leading into the forest, and it is said that if you follow it, you will be lost forever.

As you can tell, It tended to go off into storytelling mode rather than a description, so I had to cut out digressions, add in "hooks" for further elaboration ("There is also...") and ask it to try again a couple of times to produce this:

Quote from: GPT-3 & SnarkyThis painting is of a dark and spooky forest, with black trees and a red moon. There is a small path leading into the forest. There is also a small hut in the distance, with a light shining from the window. There is a feeling of something evil lurking within the painting.

I edited this a little further to give the Midjourney prompt:

Quote from: Snarkyrenaissance painting oil on canvas 2.5d of a dark and spooky forest at night, with black trees and a red moon. A small path leads into the forest. There is a small hut in the distance, with a light shining from the window. There is a feeling of something evil lurking within the painting. --ar 16:9

The style descriptions are based on the tips in this article. That produced this:



I also tried "modernist art," which also gave interesting results:

Spoiler
[close]

Then I chose one of the results to upscale to full resolution, which I in turn ran through another upscaler to get an HD image. The "oil painting" look resulted in a lot of white specks in the image (spotlight reflections in the paint), and I decided to run a "Dust & Scratches" filter on it in Photoshop to remove these. I also cropped the edges a little to get rid of "edge of canvas" artifacts. Then I scaled it down and reduced the colors to 256 to try to give it more of a game background look.
[close]

Danvzare

#39
The description of what you had to do to get that AI generated image is absolutely brilliant, and showcases just how much of you own effort you have to put in to get something worthwhile.

It's a very useful tool for amateurs like me. But it's not quite there for being wholly independent.  :-D

That being said, I don't think anyone should use it as a tool to help them make a contest entry.
Maybe to help come up with ideas, but that's about it.
Like tracing, it'll be rightfully frowned upon.

SMF spam blocked by CleanTalk