discussion How long before AI is able to make a full Renpy VN/games like the ones we play today?

[gestved]

it's hard enough to create a single image with no obvious mistakes, I don't think we are nearly that close for AI to release a half decent game complete on it's own. There is just too much room for mistakes in so many levels, even if it can create great art, it will be very hard to keep it consistent while juggling a storyline that makes sense.

 

[mistakenident7]

It'll be years out, but the day when you get AI to create a model which it remembers and records that you can then re-render in any pose with any background or clothing will be when I consider AI games something to look forward to.

 

[Gokatraz]

1 year? 5 years? 10?

Already AI is able to quickly generate compelling images, songs, and stories with minimal human involvement. I suppose a benefit of when this happens is that releases would not take so long. Or you can take unfinished games and have AI finish it with the same artwork and story style. Or even reboot ones that you didn't like the way they developed.

Any developers out there working with AI care to share their thoughts?

Objectively, you may be able to "program" one to do that right now. I've seen plenty of people add AI art into games. You'd need a couple different AI but it may be doable. I don't mess with it so I don't have a clue.

 

[noonzone]

0 to 2 years if someone bothers to program a "generator" framework, building blocks are already there
3+ for ai making that itself

 

[s3gfaul7]

Let me rephrase the question. How long before AI tools become good enough so that developers can create the completed games they want within a few weeks or days? Instead of the weeks or months it takes for just an small update? (Small gameplay wise. I'm sure the work involved is great.)

The tools for creating consistent, persistent characters are already here. It's not a prefect process yet and can get compute-intensive in certain cases, but then... so is DAZ.
Story-telling is a different matter. While 4o has made significant strides, it's still not quite there yet. Even with an experienced user who knows how to properly prompt it, the consistency is not good enough.
As for coding, 4o can already convert any dialogue to RenPy format and even query DALL-E for rudimentary images.

So, to answer your question, I believe we are not too far off that point. As it stands, AI tools can save days or even weeks, depending on workflow. And, unless something changes drastically, we keep moving forward. Every iteration of tools is an improvement, sometimes exponential. Personally, I'm quite curious to see where we are a year from now.

 

[cr503]

Once the hardware is cheep enough, I can see the AI itself being the game.

Instead of choosing from predetermined options, scene for scene, you try to find the prompts that make the AI tell the story you want.

Then you could share prompts instead of entire games.

There wouldn't be anything new the AI hasn't been trained for, so original games are still valuable.

 

[Crpapjda]

To make an entire game? probably not to long, 1year-3 tops. How long before it makes an entire game that you want to play, and cannot distinguish from a game made with human involvement? 2-5 years. How long before it makes a GOOD game, so people dont need the tag "ia generated" just to filter it because is garbage? 5 years.... unless the IA enters the cycle of consuming IA generated garbage to train, in which case, never. An IA model is just as good as the input it receives. Garbage in-garbage out

 

[HandholdDegen]

It's not going to happen, and I have no idea why we would want it to happen.
Why do we want games and entertainment to be made by an unthinking and unfeeling machine?
Machines don't understand game feel, context or mechanics. It can only replicate on a surface level without understanding the whys and hows.
If you only want visual stimulus, then grab a kaleidoscope and point it to the sun and admire.

 

[TittyCommittee]

1 year? 5 years? 10?

Already AI is able to quickly generate compelling images, songs, and stories with minimal human involvement. I suppose a benefit of when this happens is that releases would not take so long. Or you can take unfinished games and have AI finish it with the same artwork and story style. Or even reboot ones that you didn't like the way they developed.

Any developers out there working with AI care to share their thoughts?

5 years if the rate of progress remains the same.
20 years if the rate of progress halts, which it probably will because of war and stuff. These things are used in war first and foremost. so we would most likely get terminator robots first than getting AI's that can make a fully functioning game.

 

[TittyCommittee]

It's not going to happen, and I have no idea why we would want it to happen.
Why do we want games and entertainment to be made by an unthinking and unfeeling machine?
Machines don't understand game feel, context or mechanics. It can only replicate on a surface level without understanding the whys and hows.
If you only want visual stimulus, then grab a kaleidoscope and point it to the sun and admire.

Advancement doesn't care about people's feelings but yeah, it will be a pretty terrible game.

 

[EpicWaxFlame69]

I wouldn't want it. It's probably inevitable. People will find a way. If we're talking like 100% with the coding/scripting and with a coherent/consistent writing with multiple branching storylines? All AI and nothing else? Maybe 4-5 years. Like that's assuming it can store and remember all that memory.

But if we talking making AI art. AI dialogue and tweaking the prompts and scripting here and there manually? I wouldn't doubt there's already games made like that then

 

[HandholdDegen]

Advancement doesn't care about people's feelings [...]

Citation needed. What do you mean by Advancement?
AI generative technology "entrepreneurs" do not care about feelings, but they're subhuman anyway.
Advancement is just a word to describe something moving forward.

it will be a pretty terrible game.

which is why I wonder why anyone would want it.

AI game generation technology will only have two use cases: To make games that people won't make, or to make games without having to pay people for them.
Basically, it will just be used to churn out games featuring minors in lewds scenarios, because the other games it churns out will not be able to compete with actually good games made by studios.

 

[trc450]

Let me rephrase the question. How long before AI tools become good enough so that developers can create the completed games they want within a few weeks or days? Instead of the weeks or months it takes for just an small update? (Small gameplay wise. I'm sure the work involved is great.)

It is already there.

For the thread: (how-long-before-ai-is-able-to-make-a-full-renpy-game)

If you were to make a one button that makes a make game today, I think it would be possible today if it involved realistic humans only. Tools for detecting what's on the image are way ahead in realistic realm.

You would have to script the whole process into bits do them one by one as to not overwhelm the memory capacity, tech is there understand what is in the image so AI could match them together with a story. Making the story is only limited by memory(it forgets past text) or conflicting info and there is always chance of hallucination when it invents something out of thin air.

But for it to be comprehensive it would still need bug and inconsistency fixes. For it to be good and engaging, that is another thing.

 

[WillAnderson]

minimal human involvement

Nnnnno. This is not accurate.

Current AI uses ENORMOUS amounts of human involvement - more than human authors by far. They take unfathomable amounts of human constructed information and concatenate it into text, imagery, whatever we ask of them.

The problem is that current "AI" (expert systems is a more accurate description so far) simply compile what they've taken in and regurgitate over that concept. They create nothing. They concatenate, select, and copy and paste from libraries of information. This is not creation. It is my contention, though we can't (yet) verify it, that if humans do not keep creating, perhaps using these expert systems as helpful tools occasionally... nothing new will be created, and over time, "AI" will have nothing to draw upon except themselves, which will eventually result in the old, old problem from photocopier days of rendering the results terrible through copy-of-copy-of-copy-of-copy to thousands of degrees.

I happen to believe that at some point, a more general AI may possibly arise. If it does, we will see if it is any good at writing stories, painting images or composing music. But so far, computers have created no art. They've been tools of human creators at times, and they are being used to squeeze human creatives by business interests now, rendering them a relatively destructive to the very concept of artistic expression.

 

[lcflex]

Current AI uses ENORMOUS amounts of human involvement - more than human authors by far. They take unfathomable amounts of human constructed information and concatenate it into text, imagery, whatever we ask of them.

Every author, every artist, learns from the collected works of those who came before. AI training is not different.

The problem is that current "AI" (expert systems is a more accurate description so far) simply compile what they've taken in and regurgitate over that concept. They create nothing. They concatenate, select, and copy and paste from libraries of information. This is not creation.

By that definition, "creative" humans also do not create anything. 99.9999% of creativity is taking existing things and reconfiguring them into a different form. You only perceive it as "new" because nobody took those pieces and forced them into that particular shape before. This is exactly what AI does. That leftover 0.0001%? That's where the actually new ideas happen. AI isn't going to go there, not in its current form anyway. It would be a tragedy if we lost the ability to expand the space of ideas, so allowing AI to completely take over creative tasks would be terrible. That said, there's room for responsible use of AI in its current form, and I reject the argument that these are inherently not creative.

 

[HandholdDegen]

By that definition, "creative" humans also do not create anything. 99.9999% of creativity is taking existing things and reconfiguring them into a different form. [...]

No, that is not what that means. Did you think that through before you responded to it?
Please think about it a bit longer.

 

[loldude123]

I mean realistically speaking, it's possible even now. We have AI image generators and LLMs like ChatGPT, Claude and what not. Have an LLM generate a long-ass story, identify keywords for the image generation and then generate them. Lastly, put everything into code. Sure the programming is the tricky part, but all that's a relatively simple problem to solve, all things considered.

It's unlikely to happen anytime soon though because it's not worth the effort both in terms of computing power and financial reasons, because the likelyhood of the game being anywhere decent enough to reiterate on is very small.

 

[MattShizzle]

I think people underestimate. AI period can do a hell of a lot more than people would have thought even a year or so ago. In 5 or so years you could probably tell it what sort of story you want, what kinks, etc and what kind of graphics and it will spit it out. Especially if anyone comes up with something meant to do just that. AI couldn't do much of anything 5 or so years ago. In 20 or so it might be good enough it can do it easily without needing any special specific programming.

 

[SegmentCoalescing]

Probably when we reach AGI. 2030? if optimistic.
It can already help developers in some ways, but it's not advanced enough to make things by itself yet.

 

[s3gfaul7]

Probably when we reach AGI. 2030? if optimistic.

That's very, very hard to predict. Current models improve by scaling and optimizing efficiency and don't provide an evolutionary path to AGI, no matter how much computing power or raw data you throw at them. The same goes for the liquid model lines that will likely dominate in 2025/6. They will be a vast improvement over current technology, but still not on the AGI path.

No, I think when we finally get to AGI, it will come from a completely different development vector. Whether that happens a year from now, ten years or fifty - it's very hard to predict.