Vapourware How will AI affect RPGs?

[The_Sloth_Sleeps]

I'm still struggling for practical use cases.

• Art AI is just not there yet besides backgrounds/icons and other low-definition stuff.
• Dialogue writing AI is hilariously bad, can only be used for the most generic of NPCs.
• Voice AI is great, especially for narration, but hampered by the super strict licensing terms.

So basically, I'm not really using anything right now.

Also, I can almost guarantee you that once this tech becomes usable, the pitchfork crowd will want to stop devs from using it, so you'll have to put up with that.

I keep saying this. Where is the actual use case? Theres much hype but I dont see anything too exciting;
+Portrait generation (we have always had, albeit these are now more organic)
+Voice to text (we have always had)
+Text generation (we have had outsourced services produce similar blog style quality content)

Wheres the so called benefit? Besides the above whats it supposed to do exactly thats gonna change anything?

 

[Non-Edgy Gamer]

Another AI thread, another cope post by Sloth.

 

[The_Sloth_Sleeps]

Another AI thread, another cope post by Sloth.

The problem here is that you are a technology enthusiast and not capable of critically looking at the capability of these "new" technologies.
No, instead everything is "awesome" and a "game changer" or soon will be, even as we speak the technology is "advancing every day".

 

[Faarbaute]

Where I see the most immediate benefit is in asset creation. You could already buy generic assets for your game, but this will offer a lot more customisation.

 

[Non-Edgy Gamer]

Where I see the most immediate benefit is in asset creation. You could already buy generic assets for your game, but this will offer a lot more customisation.

You don't think he's been told this 100 times? Doesn't matter. Watch him complain that it's unusable and that you have no examples. Then post useable assets like portraits or loading screen art, or item icons generated with a custom model, or a how-to on creating 3d assets from generated art, or generated voice lines that match characters in old games.

It won't matter. And eventually, he'll just default back to "yeah, but it can't make seamless isometric tilesets for my game, so it's useless".

 

[The_Sloth_Sleeps]

Where I see the most immediate benefit is in asset creation. You could already buy generic assets for your game, but this will offer a lot more customisation.

You don't think he's been told this 100 times? Doesn't matter. Watch him complain that it's unusable and that you have no examples. Then post useable assets like portraits or loading screen art, or item icons generated with a custom model, or a how-to on creating 3d assets from generated art, or generated voice lines that match characters in old games.

It won't matter. Eventually, he'll just default back to "yeah, but it can't make seamless isometric tilesets for my game, so it's useless".

NEG posts character portrait: "look see wow the future".
NEG posts a synthesized voice clip: "wow AI is taking over".
Me "wheres an asset I can actually use".
NEG "the technology is advancing so quickly it will be able to do that sort of content any day now".

 

[Non-Edgy Gamer]

NEG posts character portrait: "look see wow the future".
NEG posts a synthesized voice clip: "wow AI is taking over".

The future is that this is another AI thread you're going to spam the same obviously wrong statements posed as questions over and over, and sperg out when anyone tries to answer you.

Me "wheres an asset I can actually use".
NEG "the technology is advancing so quickly it will be able to do that sort of content any day now".

More like this is the part where you revert to "b-but can it do seamless isometric tilesets?" and I say "maybe? try training it if it's that important to you." and you say "SEE I TOLD YOU IT CANT DO IT!".

It really doesn't matter what the advancement is, how many people say it's useful or how many developers say "oh, yeah, we used that for our game". You'll still come back to these threads as if it's your first post about AI and say "gosh, I really don't see the use case" before launching into some rant about how we're all tech cultists or whatever.

 

[The_Sloth_Sleeps]

NEG posts character portrait: "look see wow the future".
NEG posts a synthesized voice clip: "wow AI is taking over".

The future is that this is another AI thread you're going to spam the same obviously wrong statements posed as questions over and over, and sperg out when anyone tries to answer you.

Me "wheres an asset I can actually use".
NEG "the technology is advancing so quickly it will be able to do that sort of content any day now".

More like this is the part where you revert to "b-but can it do seamless isometric tilesets?" and I say "maybe? try training it if it's that important to you." and you say "SEE I TOLD YOU IT CANT DO IT!".

It really doesn't matter what the advancement is, how many people say it's useful or how many developers say "oh, yeah, we used that for our game". You'll still come back to these threads as if it's your first post about AI and say "gosh, I really don't see the use case" before launching into some rant about how we're all tech cultists or whatever.

Then stop jerking off to this stuff. You are making out this is an incredible game changer that everyone can benefit from. For most things, clearly its not. Worse still this is just another spam, low quality content enabler.

 

[Non-Edgy Gamer]

Then stop jerking off to this stuff.

It's useful to me, it's useful to others, devs are using it for voices and portraits already. But that's not good enough because it's not useful to you in this one particular use case you haven't even tried to train it on.

You're the guy who says a can opener is useless because it can't cook the soup too.

You are making out this is an incredible game changer that everyone can benefit from.

When did I say "everyone"? And when did you even try to benefit from it? Again, train a tileset generator. Let us know how it goes.

But you won't because you don't WANT it to be useful. You just want to spam how angry you are about AI.

 

[Dexter]

We will have AI to play the games, while people watch on YouTube or other streaming services.

https://openai.com/blog/openai-five-defeats-dota-2-world-champions/
https://venturebeat.com/ai/openais-dota-2-bot-defeated-99-4-of-players-in-public-matches/

He does have a point, though. You can quite easily prompt these image generators to reproduce a specific image from the training set

People have been finding that you can even get the AI to spit out exact replicas of current, copyrighted, works

You can't:

Spoiler

You know how you can easily reproduce a specific image using a computer? You go to Google or Bing or what have you and type the name into Image Search, then you look at it on your monitor. If you want to be particularly edgy you could mark it and hit Ctrl+C or right-click it and click "Copy" or "Save As". Hasn't really generally been much of a legal issue.

What they are doing is likely illegal within the bounds of current copyright law. However, there's grey in it that will likely need actual new laws to address what AI can and can't do when it comes to training based on copyrighted works.

UK: https://copyrightblog.kluweriplaw.c...t-and-data-mining-exception-for-all-purposes/
EU: https://mediawrites.law/the-eu-dsm-directive-new-copyright-exception-for-text-and-data-mining/
US: https://towardsdatascience.com/the-...ata-science-and-machine-learning-44cfc1c1bcaf
Japan: https://eare.eu/japan-amends-tdm-exception-copyright/

Complete, legally enforced, transparency about what derivative works and sources were used to feed AI learning models and create an artwork or piece of writing. If ChatGPT creates something for you, it needs to spit out a list of the sources it used to create it. If StableDifusion's AI creates an art work for you, it needs to spit out a list of all the artwork that it used to create it, with percentages to indicate influence.

Not quite. The fact that you can replicate specific training images from the neural network means that the original artwork is still encoded somewhere in these Gigabytes of network weights.

That's not how it works.

 

[Tyranicon]

AI tech has always been a matter of when, not if.

Its practical applications are too valuable to be left on the table.

So when I say that I'm struggling for practical use cases, I mean that today. I can't even speak for next month, and hardly next year.

The gloriously horrifying future is dawning, gentlemen.

Hold on to your codpieces, it's gonna be some weird shit.

 

[The_Sloth_Sleeps]

AI tech has always been a matter of when, not if.

Its practical applications are too valuable to be left on the table.

So when I say that I'm struggling for practical use cases, I mean that today. I can't even speak for next month, and hardly next year.

The gloriously horrifying future is dawning, gentlemen.

Hold on to your codpieces, it's gonna be some weird shit.

Oh its going to be horrifying alright, but its no way near as high tech as you would like. It will just be shit.

I can tell you what to look out for:
The internet filling up with even...more garbage.
Even more shovelware games.
More retarded chat bots instead of human service reps.
An abundance of low grade, fake information, able to be churned out quicker.
Enabling of online censorship and flagging mechanisms.

 

[Peachcurl]

AI tech has always been a matter of when, not if.

Its practical applications are too valuable to be left on the table.

So when I say that I'm struggling for practical use cases, I mean that today. I can't even speak for next month, and hardly next year.

The gloriously horrifying future is dawning, gentlemen.

Hold on to your codpieces, it's gonna be some weird shit.

I think voice generation and graphical asset generation are already useful today (though not the kind you'd use for AAA game development). Not as fully automated processes though.

ChatGPT-like text generation will not be very useful for being integrated into any RPGs soon. It may still get involved during the development process, especially by small indie devs. Say you have to cover some coding part, where the involved language is not part of your own skill set. In such a case, getting some snippets or corrections via ChatGPT is potentially useful.

 

[The_Sloth_Sleeps]

AI tech has always been a matter of when, not if.

Its practical applications are too valuable to be left on the table.

So when I say that I'm struggling for practical use cases, I mean that today. I can't even speak for next month, and hardly next year.

The gloriously horrifying future is dawning, gentlemen.

Hold on to your codpieces, it's gonna be some weird shit.

I think voice generation and graphical asset generation are already useful today (though not the kind you'd use for AAA game development). Not as fully automated processes though.

ChatGPT-like text generation will not be very useful for being integrated into any RPGs soon. It may still get involved during the development process, especially by small indie devs. Say you have to cover some coding part, where the involved language is not part of your own skill set. In such a case, getting some snippets or corrections via ChatGPT is potentially useful.

The "art" will be utilized by shovelware producers. Naturally. I cant see a wide use case outside of this.

Voice generation actually is useful, if you actually want voices for some reason. And are willing to put the effort in to writing the scripts for the voice generators. But its silly to call this AI.

ChatGPT is likely to be used as a chatbot, that you will see accompanying websites such banks and so on. When you urgently need to get hold of a utility provider to speak to a human about that huge elec bill, you will have a chatbot foisted on you. However, as far as RPGs go, no I cant see it being in common use. As for its coding ability its a downright menace.

 

[Peachcurl]

The "art" will be utilized by shovelware producers. Naturally. I cant see a wide use case outside of this.

Small teams will be able to spend more time on actually designing game systems and gameplay aspects rather than producing visual and audio assets.

Voice generation actually is useful, if you actually want voices for some reason. And are willing to put the effort in to writing the scripts for the voice generators. But its silly to call this AI.

I'm not going to discuss semantics.

As for its coding ability its a downright menace.

That's false, unless you have unreasonable expectations. It's not a 1:1 surrogate for a human programmer.

 

[The_Sloth_Sleeps]

The "art" will be utilized by shovelware producers. Naturally. I cant see a wide use case outside of this.

Small teams will be able to spend more time on actually designing game systems and gameplay aspects rather than producing visual and audio assets.

Voice generation actually is useful, if you actually want voices for some reason. And are willing to put the effort in to writing the scripts for the voice generators. But its silly to call this AI.

I'm not going to discuss semantics.

As for its coding ability its a downright menace.

That's false, unless you have unreasonable expectations. It's not a 1:1 surrogate for a human programmer.

"Say you have to cover some coding part, where the involved language is not part of your own skill set. In such a case, getting some snippets or corrections via ChatGPT is potentially useful."

 

[Peachcurl]

As for its coding ability its a downright menace.

That's false, unless you have unreasonable expectations. It's not a 1:1 surrogate for a human programmer.

"Say you have to cover some coding part, where the involved language is not part of your own skill set. In such a case, getting some snippets or corrections via ChatGPT is potentially useful."

What part do you not understand?

 

[The_Sloth_Sleeps]

As for its coding ability its a downright menace.

That's false, unless you have unreasonable expectations. It's not a 1:1 surrogate for a human programmer.

"Say you have to cover some coding part, where the involved language is not part of your own skill set. In such a case, getting some snippets or corrections via ChatGPT is potentially useful."

What part do you not understand?

The part where you are talking shit.

 

[DeepOcean]

This shit is hilarious for many reasons but I'll list just one. You do realize that this is exactly what a human being is doing when learning to craft art, do you?

Problem you are looking from the perspective of the end user, the receiver of the output, from this view, indeed the result you can get can look transformative BUT Midjourney DALL E, StableDiffusion are responsible for the input and the images they are using most have licenses where it is stated non-commercial use only. People many times get surprised when their youtube videos are taken down on copyright strikes because they thought that meme was free for commercial use when it wasnt, only because it is free from a non- commercial point of view, it doesnt mean someone didnt copyright that video/image for commercial use. Also, even if they didnt copyright, the very fact of not existing a clear license allowing use, if you use that, you are asking for trouble.

I dont see this argument that it was an Ai that made the output will convince people because the problem isnt on the output but on the input and the fact that those companies are making use of images that have licenses where it is cleary stated non-commercial use only, it is hard to avoid. I see them eliminating all the copyrighted material from their database and only using copyright free material, so they can be commercial, and I see artists in the future all copyrighting shit with very clear non Ai use clauses on it.

From the point of the view of the user, you can get away claiming transformative use so if you are making an indie game, use while it last, but from the point of view of the companies behind the input, they will have a very hard time.

 

[Maxwell's Demon]

For those who think AI will be limited to asset creation, there is a video you should watch that illustrates a far greater potential for its use in video games:



It's not a big leap to imagine this bot playing games with much reduced set of verbs, such as Counterstrike or even more complicated games like Baldur's Gate. It will be possible to train an AI that responds to the player's actions on the fly without needing to code for every possibility, and do so spending less time and money than current approaches.

 

[Moaning_Clock]

AI will be used in the most pernicious way possible. Trained to maximize profits on micro transactions. Poking every player individually to spend some cash on some shitty skin or lootbox. If this is not already done, I'm not deep into that stuff.

And then endlessly generating content for roguelikes and stuff. Warsim uses AI since a recent update afaik now.

 

[Not.AI]

The simple reality is that somebody will eventually train AI on specially ordered, properly licensed content and release that AI for free and it will be used to train other AI. Then neural network AI will be used to fill logical AI, which, unlike it, is able to explain what it's doing, why, correct errors, all the benefits John McCarthy has originally anticipated. That will also be released free and properly licensed, and then used by others to train their own special style AI that they themselves release for free, properly licensed.

More specifically, AI at that point will finally be useful and be used by developers and artists who actually have skill, as basic to train together with a couple thousand (as opposed to billion) of their own examples, and then automate, properly licensed, their own work.

However, editing and selecting the good machine-generated content from the bad or mediocre machine-generated content, as predicted by Jonathan Swift, will consume most of the time people spend creating content.

Then readers of books and editors will become the main human jobs of the future and being real heroes of socialist labor they will receive moderate rewards.

 

[Artyoan]

While I like the art generator I've been using, the images are mostly just odd. If I wanted something close to high quality I'd have to fine tune it myself, which is still the human element. I can't get it to generate good fantasy portraits which is what I was interested in. Some are okay, but things like fingers or an elf ear are apparently just not possible to do well for AI. I have no idea why. Also, random metallic looking slivers are placed on faces for no discernable reason.

I've seen really high quality big-titty cartoon/anime art being generated, which is being touched up by the obsessive men generating it, meaning its not fully AI generated to begin with.

A lot of this reminds me of crypto or NFT's. Insane hype but I'm not buying into it yet. Not idea how well a dialogue generator would work but I suspect it would have its 'fingers' issues too, and people would train themselves rapidly to discern human from machine in the writing.

If it all works out then good, we get very cheap and high quality rpgs generated out of nothing. But for now it seems like the functional nature of it is pretty limited. The only one I'm really pulling for is AI voice work though.

 

[Peachcurl]

While I like the art generator I've been using, the images are mostly just odd. If I wanted something close to high quality I'd have to fine tune it myself, which is still the human element. I can't get it to generate good fantasy portraits which is what I was interested in. Some are okay, but things like fingers or an elf ear are apparently just not possible to do well for AI. I have no idea why. Also, random metallic looking slivers are placed on faces for no discernable reason.

I've seen really high quality big-titty cartoon/anime art being generated, which is being touched up by the obsessive men generating it, meaning its not fully AI generated to begin with.

A lot of this reminds me of crypto or NFT's. Insane hype but I'm not buying into it yet. Not idea how well a dialogue generator would work but I suspect it would have its 'fingers' issues too, and people would train themselves rapidly to discern human from machine in the writing.

If it all works out then good, we get very cheap and high quality rpgs generated out of nothing. But for now it seems like the functional nature of it is pretty limited. The only one I'm really pulling for is AI voice work though.

I'm far from a professional "prompter", but just some tips:

1. Look at other people's prompts, or the AI prompt book, etc. Depending on the model you use, knowing the right order and type of keywords to use is important
2. Always start with generating a larger batch of images for your prompt, then select the ones you like (might be only a very small subset). Then:
   1. use img2img to fix what you don't like AND/OR
   2. use txt2img again, same prompt with only the seeds of the selected images, and start playing with the parameters (CFG, steps, etc.)
   3. Repeat until satisfied.
   4. Remember that you can also have negative prompts, to exclude certain features that you do not like.
3. If not successful, consider to use a different (potentially specialized) model checkpoint. For example, if you reaaaaally want anime tiddies, use AnythingV3 (or newer? Not sure if there's other versions yet.)

I agree however, that people often present cherry-picked examples. The most recent instance where I noticed this was instruct-pix2pix. The examples look great. But what I get is 100% pure crap. Doesn't compete with a slightly more invested img2img / inpaint approach.

(And also: Yes, many people touch-up their images manually via photoshop etc. Which is fair, but raises unreasonable expectations.)

 

[HammyTheFat]

I've been watching videos of dagoth ur over the last few days on youtube. They were using something from a voice AI company (that immediately shut out access after people made "racist" voiceovers of celebrities) that had you insert a few 10 second clips and then an AI could say whatever that character said. Here is my favorite one: https://www.youtube.com/watch?v=3AgaKAwEyo0

The fake videos sounded INDISTINGUISHABLE from dagoth ur in the games. It might be harder to do with lines that require emphasis or certain inflections.

I believe the future of voice acting in games is going to be companies paying someone to say a few dozen lines, then writing them a check for the right to use AI to generate all their other dialogue to save time/money. Everyone wins in this scenario.