Your thoughts on A.I. art creation

[Racoon]

So I saw this video concerning "Midjourney" an app that can create images from word prompts today:

and was pretty mindblown.

Afterwards I tried it for myself and the thing is really as good as it is shown in the video.

There is a lively discussion about what apps like these mean for the creative industry and it got me thinking about that too. Of course you can not generate the precise idea you have in your head, but I was thinking that if you are looking for something like a book or album cover or an character for your board game, it may be good enough to use this and if you want to, let an artist do some finishing touches.

I am kind of divided between thinking this is a great source of inspiration and it lets the value of art drop.

I would be very interested to know what you think about this topic :-)


By the way, an similar discussion can be applied to music as there are also AI apps being designed to cover that. I tried https://www.ampermusic.com/ and created some pretty catchy tunes with it.

[Mandle]

They are pretty amazing. There's a few next-gen ones in testing that are mind-blowing. It seems possible to join the testing, but I think you need some kind of credentials or at least a reason.

I heard that one of the biggest hurdles they are facing before release is making them "safe". They are SO realistic that they could be used to generate illegal or defamatory content quite simply, even videos that look pretty much indistinguishable from reality, and they are learning more and more and getting better all the time.

I'm actually a bit worried. Sure, they can lock them somehow to avoid questionable or despicable creations, but someone will always find a way to unlock them, or clone the technology without the safeguards in place.

We already live pretty much in a post-truth world. What's it gonna be like when someone could, say, create a video of a world leader or celebrity kicking their dog? I use a tame-ish example here because I don't wanna go too dark, but it doesn't take too much imagination to see how harmful the tech could be, if unleashed with no safety switch.

A.I. can already pretty much fake audio of a person talking so well that even expert analysis cannot distinguish it from the real thing, or at least will be able to soon. Not an expert.

But what's it gonna be like when video can also be faked perfectly?

Audio and video evidence will be very difficult, if not impossible, to use in court cases, perhaps? Will we have to go back to the days of witness testimony and such? Back to the Sherlock Holmes era of crime deduction?! Or maybe we'll just have mind-reading machines soon and court cases will become pretty much obsolete anyway?

One idea I had was: Perhaps they can embed block-chain stuff into video and audio files as proof that they are untampered originals and not fabrications? Dunno much about that either. Just an idea.

But, I'm definitely gonna be having a bit of a play with the one you mention!

[Durq]

I typed in "point-and-click adventure game background, medieval, pixel art, castle ruins in a forest".

[KyriakosCH]

Yes, AI graphic art is very impressive (wasn't aware of the music AI art, but I will check it out!)
I doubt this will extend to (non-flash fiction, mind) writing art, though. For a plethora of reasons (mostly having to do with language not being much of a tie to formal systems, which I suppose - can't be sure of course... - the AI will not be able to overcome even with the new tech it uses for pass/fail and repeat-testing)

[Racoon]

We already live pretty much in a post-truth world. What's it gonna be like when someone could, say, create a video of a world leader or celebrity kicking their dog? I use a tame-ish example here because I don't wanna go too dark, but it doesn't take too much imagination to see how harmful the tech could be, if unleashed with no safety switch.

Hey, a video of someone kicking a dog already sounds pretty horrible to me, but yes I know what you mean. By the way, when I tested the app on discord there where a lot of people using politicians for prompts, like "Putin walking into a tardis". But with the result I always felt like it did not look too simliar to the real people, maybe the app already has a restriction on that? I know it does not show explicit stuff, someone wrote something like "husband getting murdered with a knive" and the result was very vague. But it is interesting, because it shows that people are already curious about creating images like that.

Audio and video evidence will be very difficult, if not impossible, to use in court cases, perhaps? Will we have to go back to the days of witness testimony and such? Back to the Sherlock Holmes era of crime deduction?! Or maybe we'll just have mind-reading machines soon and court cases will become pretty much obsolete anyway?

That sounds pretty dystopian to me even if I would like to see some Sherlock Holmes style detective work.

But, I'm definitely gonna be having a bit of a play with the one you mention!

Have fun! I have already used up my free images and am considering a month of subscription because I am still very curious about it and have some prompts I really would like to try.

[Racoon]

I typed in "point-and-click adventure game background, medieval, pixel art, castle ruins in a forest".

That looks really cool and pretty! But I think the app might still have a problem with pixel art. I tried "pixel art dog" and the result was a pixel dog with high resolution ears. But maybe I did not try the right prompts. There are already videos about how to use the app to get better results.

[Racoon]

I doubt this will extend to (non-flash fiction, mind) writing art, though. For a plethora of reasons (mostly having to do with language not being much of a tie to formal systems, which I suppose - can't be sure of course... - the AI will not be able to overcome even with the new tech it uses for pass/fail and repeat-testing)

I was thinking about this too. While it will definitely be a challenge to make something like a AI writer, would it not be possible to feed it all kind of ebooks and make connections between how they are constructed? Then similar as with the art app you would prompt something like "scary, witch, futuristic" and would get a randomly created story.

[KyriakosCH]

I doubt this will extend to (non-flash fiction, mind) writing art, though. For a plethora of reasons (mostly having to do with language not being much of a tie to formal systems, which I suppose - can't be sure of course... - the AI will not be able to overcome even with the new tech it uses for pass/fail and repeat-testing)

I was thinking about this too. While it will definitely be a challenge to make something like a AI writer, would it not be possible to feed it all kind of ebooks and make connections between how they are constructed? Then similar as with the art app you would prompt something like "scary, witch, futuristic" and would get a randomly created story.

It might, but I was wondering if any connections the AI will "make" can fundamentally be unlike the connections you see in formal logic, which (given the output is just natural language) may be not enough to present anything workable past a very short word limit.

[Snarky]

I mean, already a couple of years ago we were playing AI Dungeon, which is basically a collaborative AI story-writing tool. You can read some examples of stories written by OpenAI's more advanced GPT-3 system here.

You can see that the current generation of AIs can produce text that reads just fine on a sentence-by-sentence, paragraph-by-paragraph level, and usually remains meaningful and relevant to the given topic. However, it definitely still struggles to maintain coherence and focus over longer stretches. Its stories tend to be rambling and pointless. But it IMO the performance of text-producing AIs is comparable to that of image-generating AIs.

There's also this. (The text in green was written by GPT-3, while the text in white was supplied by the person interacting with it, @funnycats22 on Twitter):

[Kastchey]

Holy cow. That story makes so much sense it's incredible it was almost entirely written by an AI.

[Racoon]

I mean, already a couple of years ago we were playing AI Dungeon, which is basically a collaborative AI story-writing tool.

Oh, I really have to check that out, sounds interesting.


I googled AI writing and found this site https://narrative-device.herokuapp.com/, where you can give two prompts and generate a story. When you give your email adress you have five free tries. So here are my three stories about sunflowers and tradgedy:

Spoiler

[close]

They are pretty short and you can guess that they are not written by a human mind, but it has potential I think.

By the way, the first one is my favorite!

[Kyrridas]

of course there are bigger implications surrounding the technology and its place in modern society, but as far as gaming/development goes....yeah, i dunno. it could theoretically put some concept artists out of a job, i guess. big studios dont bother making pixel art anymore. and while theres probably an AI out there that could whip up some 3d character models, it probably wouldnt animate them too well. so your developers are still going to have to work with the models, so they may as well create them. basically, i dont think the gaming industry has to be afraid of it until we reach the point of "type in a text prompt and auto-generate a fully playable game"....which might not be _too_ far off...?

i will say: i have near-zero artistic ability and no money to commission anything. so ive been using some AI image generation to help create character portraits for my current AGS project.

[Danvzare]

They can make some pretty good stuff. Enough to impress even me, which is saying something considering I'm usually rather disappointed with the results when something like this comes along.
(Also, it pairs well with AI scaling tools, just so you know.  )

The thing is though, while most people are saying things like "Could this replace artists?" or "The dangers of using this!" I'm instead thinking "How can I use this as a tool to aid me?"

For example, it'd be brilliant if I could take a photograph and then have it be redrawn in the style of The Curse of Monkey Island. That's not going to happen though.
But it turns out that this particular tool is brilliant for giving you that initial spark of inspiration, like what pose to use or how to angle a shot. It's never going to output something I could actually just straight up use, like a background for a game, or a walking animation. But it could give me something that's a little more difficult to get than simply drawing it, like a picture of an open blank book or a blank piece of paper. Something that you could actually edit and use ingame, and not have to worry about taking a picture yourself or searching for a stock photo online.



Now can anyone point me towards an AI where I can feed in the script of every single Slasher movie so I can see what it spits out. The thing about Slasher movies is that they're all basically the same movie, just with a few slight alterations. It's actually why I like them. It's fun to see what part of the formula they decide to keep and what they decide to change. And I really want to see what an AI would consider to be the most generic Slasher movie of all time!

[Mandle]

The thing is though, while most people are saying things like "Could this replace artists?" or "The dangers of using this!" I'm instead thinking "How can I use this as a tool to aid me?"

I didn't mean to infer that I'm only worried about this tech. Your final point is a given. It is what the tool is intended for. I was just going beyond my excitement and amazement at the thing.

Now can anyone point me towards an AI where I can feed in the script of every single Slasher movie so I can see what it spits out. The thing about Slasher movies is that they're all basically the same movie, just with a few slight alterations. It's actually why I like them. It's fun to see what part of the formula they decide to keep and what they decide to change. And I really want to see what an AI would consider to be the most generic Slasher movie of all time!

One of the next-gen versions I saw was used to create an entire graphic novel with a storyline only hinted at through the artist's initial prompts. Of course, it still had to be adjusted here and there by the artist, but even the first images and page-layouts were incredible. And these things are only gonna get smarter and smarter at their job once the entire planet is using them. An amazing wild ride ahead indeed!

[Babar]

For example, it'd be brilliant if I could take a photograph and then have it be redrawn in the style of The Curse of Monkey Island. That's not going to happen though.

Pretty sure AI tools that take in 1 image as input for the style/aesthetic and the other image as input to convert to that style/aesthetic already exist.

[KyriakosCH]

of course there are bigger implications surrounding the technology and its place in modern society, but as far as gaming/development goes....yeah, i dunno. it could theoretically put some concept artists out of a job, i guess. big studios dont bother making pixel art anymore. and while theres probably an AI out there that could whip up some 3d character models, it probably wouldnt animate them too well. so your developers are still going to have to work with the models, so they may as well create them. basically, i dont think the gaming industry has to be afraid of it until we reach the point of "type in a text prompt and auto-generate a fully playable game"....which might not be _too_ far off...?

i will say: i have near-zero artistic ability and no money to commission anything. so ive been using some AI image generation to help create character portraits for my current AGS project.

I think it could, very realistically, be used for indie computer games. At some point (I wouldn't be surprised if it is here already) you will have AI doing bulk animation too (in most games, there is a specific set of generic animations and the rest is a change of skin, so it is already mass-produced by an iteration). If the AI can generate the first animation (or a few types of those), and also the skins, that's all.
Copyright may be a serious problem there. I suppose this is why you cannot buy a version of the AI and have to use it online/be part of a program (so can always be traced back).

Of course it is very bad for people who want to make their own graphics/animations. And indeed, the AI won't fully express your vision. But it can easily surprise you with something that is also inspirational, and so to your liking.

[KyriakosCH]

Saw this article posted in another forum I go to:

"Meta's chatbot says the company 'exploits people'​
Meta's new prototype chatbot has told the BBC that Mark Zuckerberg exploits its users for money.
Meta says the chatbot uses artificial intelligence and can chat on "nearly any topic".
Asked what the chatbot thought of the company's CEO and founder, it replied "our country is divided and he didn't help that at all".
Meta said the chatbot was a prototype and might produce rude or offensive answers.
"Everyone who uses Blender Bot is required to acknowledge they understand it's for research and entertainment purposes only, that it can make untrue or offensive statements, and that they agree to not intentionally trigger the bot to make offensive statements," said a Meta spokesperson.
The chatbot, called BlenderBot 3, was released to the public on Friday.

The programme "learns" from large amounts of publicly available language data.
When asked about Mark Zuckerberg, the chatbot told the BBC: "He did a terrible job at testifying before congress. It makes me concerned about our country."

Mr Zuckerberg has been questioned several times by US politicians, most notably in 2018.

"Our country is divided, and he didn't help with that at all," the chatbot continued.
"His company exploits people for money and he doesn't care. It needs to stop!" it said.
https://www.bbc.com/news/technology-62497674"

Well, there you have it. If that chatbot actually was AI (had intelligence) it would never have said something against its owner 

[Racoon]

The thing is though, while most people are saying things like "Could this replace artists?" or "The dangers of using this!" I'm instead thinking "How can I use this as a tool to aid me?"

I am thinking all of this. For now it is a great inspiration and can make an artists life easier when it comes to finding colour choices or composition ideas etc. But as it is getting better, like all technology, I think it could become a valid concern that it can replace artist to some extend. A lot of artists also make money with social media like with posting new artworks on instgram and creating a following. Now instgram is already getting flodded with midjourney art that was done in seconds and (almost) looks like done by a professional. Does that create a false picture of how much work a "real" artwork takes?

For example, it'd be brilliant if I could take a photograph and then have it be redrawn in the style of The Curse of Monkey Island. That's not going to happen though.

Pretty sure AI tools that take in 1 image as input for the style/aesthetic and the other image as input to convert to that style/aesthetic already exist.

I think I also read about something like that being possible with midjourney, but it does not work that well because the developers are scared what pictures might be used..but I cant find the source where I saw that again.

[Danvzare]

For example, it'd be brilliant if I could take a photograph and then have it be redrawn in the style of The Curse of Monkey Island. That's not going to happen though.

Pretty sure AI tools that take in 1 image as input for the style/aesthetic and the other image as input to convert to that style/aesthetic already exist.

They do, and they're not good. There was a thread on here hyping them up, and I showed how terrible my results were.
It turns out that they're good if you want useless abstract art. Which are pretty to look at, but useless for game making. But if you want something like what I just said "a photograph and then have it be redrawn in the style of The Curse of Monkey Island" I don't think that'll happen anytime soon. (Try it yourself if you don't believe me, you'll just get an ugly collage.)

The thing is though, while most people are saying things like "Could this replace artists?" or "The dangers of using this!" I'm instead thinking "How can I use this as a tool to aid me?"

I am thinking all of this. For now it is a great inspiration and can make an artists life easier when it comes to finding colour choices or composition ideas etc. But as it is getting better, like all technology, I think it could become a valid concern that it can replace artist to some extend. A lot of artists also make money with social media like with posting new artworks on instgram and creating a following. Now instgram is already getting flodded with midjourney art that was done in seconds and (almost) looks like done by a professional. Does that create a false picture of how much work a "real" artwork takes?

I doubt it'll replace good artists. Just the amateur ones that think they're special (which there are way too many of in my opinion).
After all, it'll be a very long time before these programs can make something which could be used without any input from an artist. Making you that perfect background or animation. And until that point in time (if it ever happens), people will have to go to an artist. And if they aren't willing to pay the price, then they'll have to do it themselves, perhaps even using that tool. In turn letting them gain an appriciation of how to make art themselves. Worst case scenario, we get another huge influx of amateur artists with huge egos.

And does it create a false picture of how much work real artwork takes? Yeah, probably. But people already have that false picture. You already get people wanting to pay artists in exposure due to not realizing this. And there are many MANY people who don't want to pay for anything because they think it's not worth as much as it actually is, and it's prevalent in any field. Ask any gardener about it, and you'll hear just as many horror stories about it as you would from an artist.

Saying this will replace artists is like saying how the camera replaced painters. Because what's the point of buying a painting of a landscape when you can take a photograph of it? It's like saying how digital painting software like photoshop replaced traditional artists, because why paint when you can easily edit, copy, paste, and print your artwork?
On paper, it sounds like something that should be worrying. But in reality, it doesn't make any sense. And that's because you're worried about something being replaced by something that only appears to be equivalent on the surface.

It's a paradigm shift, yes. But it's not going to make an entire field obsolete. We could literally invent the food synthesizers from Star Trek, and the culinary industry would still thrive. Artists will be fine. 

[Snarky]

They do, and they're not good. There was a thread on here hyping them up, and I showed how terrible my results were.
It turns out that they're good if you want useless abstract art. Which are pretty to look at, but useless for game making. But if you want something like what I just said "a photograph and then have it be redrawn in the style of The Curse of Monkey Island" I don't think that'll happen anytime soon. (Try it yourself if you don't believe me, you'll just get an ugly collage.)

I don't think that's a very accurate summary of what happened. (Edit: Actually, it was this thread.) You had a very specific thing you wanted it to do, and it turns out it couldn't do that particular thing.

It works much better with traditional realistic/impressionistic painting (as well as abstract stylization), probably because that is what it has been trained to do:





All these deep neural network tools rely heavily on their training data. So if you trained it on a lot of cartoony art, it could probably do a half-decent job at turning photos in COMI-style paintings, too.

But I think that's really the crux: It's fairly easy to get it to do something that looks pretty good, if you're just messing around and aren't aiming for anything in particular. But in my experience, having played with a number of these tools (Neural Network filters are actually implemented in Photoshop now, including artist style transfer), if you do have something specific in mind, it's very hard to get it to do what you want, and it almost never looks right.*

So I mostly agree that it's not currently a serious threat to most worthwhile artists/illustrators. That could change, though.

* 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).