A pattern I’ve found useful in other settings is starting with code for an existing “game” that sort of resembles what you want to make and then modifying components until you have a whole new game but it shares similar infrastructure to the original. So you benefit from the existing system and avoid a lot of problems.
That's the part that is visible to everyone else, so that's the part that an LLM can see. That means someone else can clone your ideas and aesthetics. The ol' double-edged sword.
Absolute nonsense. There is no cookie-cutter formula for game development, despite what the idea guys think. Mario 64 was literally just Mario in a single small room until the last six months of development. They focused purely on the actual game - the mechanics, physics, movesets, items, and enemies - inside that tiny room until it felt perfect. Then, in one final push, they quickly slapped together all the levels and the storyline at the very end.
We are very quickly going to see the reality: games are made by programmers. Everyone else who has spent decades comfortably riding on the programmers backs is being replaced by AI. This includes the artists, the junior programmers, the testers, all of them. The future isn't bloated teams, it's going to be just one lone genius (or a small group of programming geniuses banding together) and his magnum opus across the board.
If the Touhou games or Cave Story were released today, all of Hackernews would be like "dude, I wonder what their LLM workflow is like!" Japanese solo hikikomori devs have been putting out insane stuff since long before LLMs emerged.
Not really, those games are very simple code wise. A high schooler could do it (source me).
You could make a bullet hell game engine as a project in an intro CS course.
The hard part is the content in the game, and ZUN was already a composer. That just leaves the code which is easy, and the bullet patterns, which ZUN clearly improved at through his earlier games. (and the art, which is famously bad though endearing)
The other comment said that obviously no solo dev do everything by himself. They must have asset maker or song maker who do all things mostly uncredited. But, here we are, of course there are true solo devs!
> Not really, those games are very simple code wise. A high schooler could do it (source me).
That very much depends on how much they did themselves. If they used unity, and went very light on the simulation, sure.
> You could make a bullet hell game engine as a project in an intro CS course.
No you couldn’t. Well you could but it wouldn’t be appropriate for actual beginners unless you stripped it down so much that calling it an engine was meaningless.
> No you couldn’t. Well you could but it wouldn’t be appropriate for actual beginners unless you stripped it down so much that calling it an engine was meaningless.
You definitely can. One of the assignments in the CS intro course I took was a bullet hell game. "calling it an engine was meaningless" is an opinion that requires ignoring the fundamentals of what a game engine is.
A game engine is a framework that allows you to create games. Assuming you don’t have tons of content (and I’m sure you didn’t in an intro CS class), building a game is an easier task than building a game engine.
Let me ask you this. What were the parameters of your assignment? What libraries were you allowed to use.
By modern standards, yes, writing a bullet hell shooter game is not hard.
But ZUN started on the PC-98.
To put that platform in a western context, imagine if IBM had gone with planar graphics for VGA. Or an Amiga with no coprocessors, sprites, or scrolling[0]. You have a lot of pixels to fill and no help to do it with. It can't even run DooM[1]. Most other developers threw their hands up and shipped RPGs, erotic visual novels, or porn. Getting a fast action game running on PC-98 is a genuine accomplishment.
[0] I am aware that I just described a compact Macintosh.
It actually seems to be a relatively small vocal group. I've marked most of them red (as I previously did the one above) with https://hackersmacker.org
I’m wondering the same thing. I’ve been thinking about getting into solo LLM game dev. I don’t know the first thing about it
You’re all set!
A pattern I’ve found useful in other settings is starting with code for an existing “game” that sort of resembles what you want to make and then modifying components until you have a whole new game but it shares similar infrastructure to the original. So you benefit from the existing system and avoid a lot of problems.
What would be your added value?
The ideas and aesthetics
That's the part that is visible to everyone else, so that's the part that an LLM can see. That means someone else can clone your ideas and aesthetics. The ol' double-edged sword.
It only works when starting with open source to show to LLM. To monetize my modification, I would not make mine open source.
If the licensing allows for it I’m fine.
If the LLM does the work, why should someone pay you when they could use their own tokens?
Everything. The story, how the game works, the art assets, voice lines, music, etc.
Programming is like 10% of a game. The world building and UX is the juice.
Anybody can make Lord of the Rings, but there's only one.
Absolute nonsense. There is no cookie-cutter formula for game development, despite what the idea guys think. Mario 64 was literally just Mario in a single small room until the last six months of development. They focused purely on the actual game - the mechanics, physics, movesets, items, and enemies - inside that tiny room until it felt perfect. Then, in one final push, they quickly slapped together all the levels and the storyline at the very end.
We are very quickly going to see the reality: games are made by programmers. Everyone else who has spent decades comfortably riding on the programmers backs is being replaced by AI. This includes the artists, the junior programmers, the testers, all of them. The future isn't bloated teams, it's going to be just one lone genius (or a small group of programming geniuses banding together) and his magnum opus across the board.
Most of my favorite games don't have physics engines, and many are story sequels. One of my favorites is the modern Equivalent to Myst.
Good luck with your idea tho
I vouched for you
I'd be happy to help you! I'm working on a game myself.
My first piece of advice is: Pick one mechanic or idea, and ship it all the way to a player (a friend) to see if it's legible or fun.
Thanks for the offer! Unfortunately I don’t have the time right now, but I am going over some ideas at a high level.
Are you building a single player or multiplayer game?
Step 1: acquire land for datacenter.
If the Touhou games or Cave Story were released today, all of Hackernews would be like "dude, I wonder what their LLM workflow is like!" Japanese solo hikikomori devs have been putting out insane stuff since long before LLMs emerged.
"Please respond to the strongest plausible interpretation of what someone says, not a weaker one that's easier to criticize. Assume good faith."
"Don't be snarky."
"Please don't sneer, including at the rest of the community." It's reliably a marker of bad comments and worse threads.
https://news.ycombinator.com/newsguidelines.html
Not really, those games are very simple code wise. A high schooler could do it (source me).
You could make a bullet hell game engine as a project in an intro CS course.
The hard part is the content in the game, and ZUN was already a composer. That just leaves the code which is easy, and the bullet patterns, which ZUN clearly improved at through his earlier games. (and the art, which is famously bad though endearing)
The other comment said that obviously no solo dev do everything by himself. They must have asset maker or song maker who do all things mostly uncredited. But, here we are, of course there are true solo devs!
> Not really, those games are very simple code wise. A high schooler could do it (source me).
That very much depends on how much they did themselves. If they used unity, and went very light on the simulation, sure.
> You could make a bullet hell game engine as a project in an intro CS course.
No you couldn’t. Well you could but it wouldn’t be appropriate for actual beginners unless you stripped it down so much that calling it an engine was meaningless.
> No you couldn’t. Well you could but it wouldn’t be appropriate for actual beginners unless you stripped it down so much that calling it an engine was meaningless.
You definitely can. One of the assignments in the CS intro course I took was a bullet hell game. "calling it an engine was meaningless" is an opinion that requires ignoring the fundamentals of what a game engine is.
A game engine is a framework that allows you to create games. Assuming you don’t have tons of content (and I’m sure you didn’t in an intro CS class), building a game is an easier task than building a game engine.
Let me ask you this. What were the parameters of your assignment? What libraries were you allowed to use.
By modern standards, yes, writing a bullet hell shooter game is not hard.
But ZUN started on the PC-98.
To put that platform in a western context, imagine if IBM had gone with planar graphics for VGA. Or an Amiga with no coprocessors, sprites, or scrolling[0]. You have a lot of pixels to fill and no help to do it with. It can't even run DooM[1]. Most other developers threw their hands up and shipped RPGs, erotic visual novels, or porn. Getting a fast action game running on PC-98 is a genuine accomplishment.
[0] I am aware that I just described a compact Macintosh.
[1] https://www.youtube.com/watch?v=Fj0-KvV0SC0
What you described is even better: an Atari ST :)
the (physical) zuiki mascon seems like a labor of love too.
This place has become an AI-focused hellscape. It really is sad.
It actually seems to be a relatively small vocal group. I've marked most of them red (as I previously did the one above) with https://hackersmacker.org
The people touting LLMs or the people complaining about LLMs? :)
Thanks for the recc, very cool extension!
I'll sometimes search for the LLM string on random non-AI related articles. It's sad.
[flagged]
I wish this was true. That AI slop couldn't reach prod and polute our virtual stores and assets marketplaces.
Pretty much nothing has shipped without LLM involvement over the last 6-12 months