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

[kazu13]

Do you mean you just input a prompt and makes the whole game for you including cg,script, bgm etc? Then it might take 3-5 years at max.

But if you're talking about making them separately and combine those to make one then it can be done now. But you may need a powerful computer to generate consistent character cgs with ai.

 

[leomgr69]

10 years from now 100% for sure 5 years maybe 30% 1 year probably no chance.

 

[s3gfaul7]

This is unlikely to ever occur for many reasons, but the simple way to explain it is that "AI" is basically just a word salad machine and then humans put sorting bins under the word salad bowl and code and recode the robot's algorithms until it mostly sorts correctly.

Someone should probably phone the

You must be registered to see links

and tell that to the President and his advisors. The US government seems confused on the subject and worries for nothing.

 

[odspyykt]

Do you mean you just input a prompt and makes the whole game for you including cg,script, bgm etc? Then it might take 3-5 years at max.

But if you're talking about making them separately and combine those to make one then it can be done now. But you may need a powerful computer to generate consistent character cgs with ai.

I mean AI tools in general helping in any way possible to speed up development to weekly or monthly updates instead of yearly. (eg: image creation, story line progression, dialogue, automation, etc). In fact I've noticed since posting my question, I've seen a number of games come out (version 1.0) completely finished already but with very little or no comment history nor reviews. Haven't downloaded any of them, but I'm guessing AI has had a hand.

You must be registered to see attachments

 

[herelooking027]

Someone should probably phone the

You must be registered to see links

and tell that to the President and his advisors. The US government seems confused on the subject and worries for nothing.

Actually, it's kind of the opposite. Because AI is a word salad machine, misconceptions that it provides correct results when told to do something is incredibly dangerous. Not just for "self-driving" vehicles, but for production and spread of incorrect information, disruption of security systems, and so on. And what "AI" does do well (which is showing people examples of what other people have done before), it is exposing information that may have been intended to be kept private: from social security numbers, to code that runs the energy grid, to national security secrets.

AI needs massive regulation immediately, but no one in government who has the power to do that has any clue about the real dangers, much less how to safeguard against them.

 

[upbutte]

To make a good AVN, 5-10 years. To make 92% of the AVNs currently being made, less than 5 years.

 

[KantamaroSan]

Do you mean you just input a prompt and makes the whole game for you including cg,script, bgm etc? Then it might take 3-5 years at max.

But if you're talking about making them separately and combine those to make one then it can be done now. But you may need a powerful computer to generate consistent character cgs with ai.

I've played around with generative AI, and my gut feeling says you're wrong...
I've tried varies language models, none of them could write an internally consistent engaging story, let alone a scenario for a VN where choices impact branches. LLM does not seem to understand what it's writing.
I've played with code generation, and if the VN is a run-of-the-mill choices, branches and text, I think this is indeed doable, it's relatively consistent and simple code across different VNs.
I've played with image generation, and the amount of prompt correcting and 'engineering' you have to do to get what you want is just stupid, I admit I am inexperienced, and a lot of people can do this better, but you also need consistent characters and locations, consistent anatomy, consistent sizes over the entirety of the VN images, and that's just not feasible right now with the off-the-shelf tools.

And then just combining these? Well maybe if you're ok with low quality. And I'm also not doubting that it's possible to generate a 'decent' base of a game, just... it's going to need a lot of extra work on top of the one prompt. I do not believe we'll make enough progress to make that possible in the next five years either. Sure there will be improvements, sure we'll be able to AI generate faster and more. But we won't get to a point that it's indistinguishable from a human produced work within 5 years. I hope I'm wrong, I hope I'm impressed in 5 years about what I can do with a single prompt... For now, put me on the disbelievers list.

 

[herelooking027]

I've played around with generative AI, and my gut feeling says you're wrong...
I've tried varies language models, none of them could write an internally consistent engaging story, let alone a scenario for a VN where choices impact branches. LLM does not seem to understand what it's writing.
I've played with code generation, and if the VN is a run-of-the-mill choices, branches and text, I think this is indeed doable, it's relatively consistent and simple code across different VNs.
I've played with image generation, and the amount of prompt correcting and 'engineering' you have to do to get what you want is just stupid, I admit I am inexperienced, and a lot of people can do this better, but you also need consistent characters and locations, consistent anatomy, consistent sizes over the entirety of the VN images, and that's just not feasible right now with the off-the-shelf tools.

And then just combining these? Well maybe if you're ok with low quality. And I'm also not doubting that it's possible to generate a 'decent' base of a game, just... it's going to need a lot of extra work on top of the one prompt. I do not believe we'll make enough progress to make that possible in the next five years either. Sure there will be improvements, sure we'll be able to AI generate faster and more. But we won't get to a point that it's indistinguishable from a human produced work within 5 years. I hope I'm wrong, I hope I'm impressed in 5 years about what I can do with a single prompt... For now, put me on the disbelievers list.

Most people who don't work in the field or don't do hobby work are ignorant that LLMs don't remember more than a few thousand words of text at a time if you have a very powerful computer. Should we let them dream and let them be scammed by liars who say their game runs on AI, or tell them truth and let them be sad before they get scammed?

 

[indipiccalilli]

AI is a con, just like the dot com bubble. I doubt this would be possible or it will be unplayable garbage.

 

[DarEx947]

Without any Humans... along time. but there are several good games out there that are made using AI. Puzzle Pieced together.

 

[wintercoat]

With a shit, nonsensical story, terrible scene composition, and ugly characters? Could probably whip something up now if you put the effort in. An actual good AVN with a decent story, good looking and varied characters, and enjoyable sex scenes? I'd say maybe 8-10 years if someone were to put in the effort curating an AI starting now, 15-20 years for paid tools to be refined enough for someone to whip up a game in a few months.

 

[abarosant]

A decent one? 10 years and with close supervision

 

[Monseigneur]

If I were Elon Musk, I would promise that in two years the Renpy games would write themselves. But in fact, no one knows for sure for 2 years or 20 years. We can only guess.

 

[PervyGod]

Current AIs are over hyped. Best description for them i've heard is over glorified chatbots. So it's going to be hot while before AI can good games for us.

 

[ChaosLoli]

It's going to be a while IMO. The current AI techniques are just remixing old stuff, and training them on something as complicated as a full game would be insanely hard. Not to mention that there isn't a ton of good data out there for them.

 

[broooooooxddfd]

mmmhhhh, I think that the problem rn is about consistency, and that's a gap that will take years till doesn't feel like a problem