"You know, you should also build me a FPS. Insanely detailed quality pls, something that looks visually like it can keep up with AAA. Make it small in scope (like a tiny level) but otherwise impressive photorealistic graphics"
One perspective on the big AI corps is that they’re a grand experiment in capitalism: given a letter of marque to maximally exploit the combined intellectual capital of the world to date, what’s the most you can make? It is very unfortunate (although arguably also very predictable) that the answer is “mostly a bunch of hacks and ripoffs”.
Fine, so AI can make generic shooters. But the interesting thing about games is not graphics, but engaging mechanics. Like enemy AI, not in the sense of a chess engine destroying you, but with a theory of mind that guesses where you are and what you're doing, so you can manipulate and outsmart them.
Interesting games are the ones where you can hack your way into places where you "are not supposed to be," and games where you finish and think, "I could have done it better," or "What would have happened if I had done it differently?"
I've been building an FPS since I realized Opus 4.8 claude had some essence of taste and modeling capability. I have 0 experience in game development. Along the way I learned how hard video game development really is and have a new respect for indie studios. AI will never replace original story telling and artwork - having a firm foundation and building the first models and level myself was key.
Since then, each model advancement has made it easier and easier to delegate and form workflows around each aspect of making my game - to the point that I login once and awhile to sculpt, get visual feedback for textures lighting and level design... I get a level done a week - test it myself, and ship it to beta testers who auto update and can send me anonymous feedback - all thanks to Opus.
I'm one dude - and I think this is a crippling vision to some of what's to come. AAA studios certainly know and will serve you AI generated games instead of paying talented people. I hope those talented people getting displaced can get enabled cause what I've done is pretty amazing.
AAA slop has been going on for years before ai era where slop become a word. It's not about mass production slops, it's about making something that genuinely sells.
Not saying you are producing slops. Just look at what truly sells on steam chart. The game industry is a creative industry just like others. Anyone can bang together some music or videos and upload them, but which ones are really getting popular and making money? That's the real question
When my all in stake is about, oh, say $5k - how can you afford not to try? Making it good is tough, I've been at this for nearly half of a year and I'm only now hitting my stride ( a level per week )
Especially if it's just gratifying to say you released something you're proud of. I have no illusion AAA titles with big budgets/advertising budgets and higher polish will give it a slim chance but,
What are you talking about? Even charitable read of the current FPS scene shows biggest names rehashing the same stories and worrying more about presentation that anything halfway interesting. AAA is over? Good riddance. It is not over but AAA studios will just send slop?
Dude. They are already doing it. The only difference now is that you can too.
Hopefuly, I gave you an idea of what you need to do.
Sure, they've been sending more slop lately but at least there was some aspect of providing a livelihood for game devs. It sounded like that was the comment was expressing anxiety toward, at least how I read it. I guess game devs will just have to shift to something else, but as a (non-game) dev from the before times it's really not clear where to shift that energy. Especially if you used to depend on it to provide all your essentials for life.
I hear you and I don't really have a solution. However, what I suspect is that status quo can't go on as-is anyway. I can't say I am not worried about what is on the horizon, but just propping it all up so that company can release Modern Warfare MM seems wrong on a different level.
I guess my question is whether it actually modeled the assets, or simply pulled a bunch of cc0 assets off the internet.
This has actually be a challenge for my agentic gamedev efforts. Getting to a particular visual style outside of PBR assets pulled off of packs online is more challenging right now.
This isn't a completely original approach: for example kkrieger did this some years ago as part of creating an extremely small FPS. It is not too surprising that it's something that is more suited towards LLMs (and vice-versa).
> Everything in this game is from scratch: models, sounds, textures. ~$550 at API pricing, ~2.7M output and 445M input tokens, heavily cached. Took 8 hours!
How is the discussion here so negative? I am frankly astonished that this was possible to generate, even with human guidance. No idea why everyone is in such a rush to dismiss it as a reskin, or complain that it doesn't work well in mobile. Most video games i have interacted with that look like this don't work in mobile at all???
The more examples there are available for something, the less words you need to describe it as it can just infer the gaps and come up with something that seems OK.
Ran out of rifle ammo, died when my small arms fire couldn't hold off the drones. Realized I can't see them that well but they can all see and pew me pretty easily. Really says a lot about where we are heading.
note: hit a bug on bog standard chrome where I cannot toggle between "mission" and "holdout" once I click holdout. Is it a commentary on where we're heading too? Was this formally proven to follow a spec during design?
It does seem, across a range of fields, that the main benefit of AI is that it decreases the cost/time required to produce dreck: documents that no one reads, websites that no one visits, games that no one wants to play.
That's a lucrative technology, sure, but I'm yet to be convinced it's a useful one. We already had more lazy asset flips than we wanted.
I believe it drastically decreases the cost/time required to produce stuff. Some people are picky about the specifics/quality of produced stuff, some aren't, and some are welcoming specifically the quantity, for less than savory reasons.
It's a terrible game in the same way 10 years ago asset flips that used public tutorial blueprints and stolen assets from the unreal store were terrible games. We properly pointed out how big of a waste of time and how "unethical" for lack of a better word these games were, and we should do the same here.
A keen eye will also notice the immense amount of small random objects clipping into each other, the lack of culling, the at-first-glance models making sense unless you sit still to study them, such as the floating stairs directly above the player in the twitter clip mentioned in another comment.
Maybe it's incredible that a model can come up with something like this, but I find the spam of low quality slop with no direction or human input behind them to be kind of disrespectful to everybody's time. Who are these experiences for? Other than giving Anthropic $500 cause you're bored.
Maybe I'm biased because social game dev circles have been completely ruined by AI posters shitting everything up.
I played it and it was fun, I absolutely felt suspense and performance was very good considering it's a web game. Music was fitting but still quite bland. The atmosphere looks good, but was obviously designed to show off advanced graphics...
I like how there were hiding spots and how the bullets of the enemies can reflect off the terrain, allowing you to find hideouts and gain height leverage to take out ground enemies. it's hard to determine the height of drones but could be intentional. I also like how the game certainly isn't too easy. Unfortunately the keybinds aren't rebindable, which is a shame because that means I can't use the visor on a touchpad. Also the undersides of some of the stacked shipping containers aren't properly rendered. Drones clip through some the containers in the distance, probably because they aren't there and part of the skybox, and drones interact weird when colliding with containers quickly jumping up and down to compensate.
7/10 please never generate something like this again.
I really don't get the religious fervor here. Artists can now focus on what is the actual ( and claimed by, sometimes, the same artists ) important thing: gameplay. Isn't that a great thing that we can get some of the crud out of the way? And if you want to do handcrafted engine where snow is generated as proper fractals, it is still an option.
But that is not the real objection is there? The bar has simply been raised.
Edit: FWIW, stories in games were getting away with murder so it is about time that changed.
Edit2: Geez man. Talk about apt game title too given the response.
High-quality games come from people caring deeply about every single aspect; I just don't believe there's any reason to think that someone phoning in most of it will put sufficient care into the aspects they do themselves, or even notice the lack of cohesion between elements.
wooow. this would be easily several months of work 2 years ago.
you could easily put some idea behind, put donald t there, or something trendy now, and get at least some attention/fun, and maybe some money on ads . the capability is there.
The value was the joy in creating it. We shouldn’t pass judgement on other people for things like this. I don’t enjoy collecting coins or bird watching, but don’t mind if other people do. The amount of money spent creating it changes the conversation a bit, but people spend money on trivial things with no perceived value.
We’re moving into a new era where AI is used by a certain range of people (computer enthusiasts) like the earlier days of the internet. Wait until grandma starts using AI like when everyone started using the internet. Look at how that turned out. We’re not there yet, but that’s what you need to worry about.
> the amount of money spent creating it changes the conversation a bit
Both coin collectors and and bird watchers (and members of any other hobby) would look askance at someone who poured money into the hobby but didn't demonstrate passion otherwise.
Two common ideas in our culture are that [resource] is not a substitute for passion, and that buying something is not the same as creating/earning/deserving it. These aren't new ideas, and they don't suddenly disappear because AI is here.
> where AI is used by a certain range of people (computer enthusiasts)
I'm not sure that this is the case; I think AI is used by AI enthusiasts, and a lot of the pushback is because they're joining [thing] enthusiast communities without actually caring about [thing] or the community.
No, it isn't. With $550 you could buy ten really good games made by someone who knows how to make games, or spend it on something else you enjoy or invest it. If he wasn't a manager at Anthropic this would be $550 of money wasted. Nobody will enjoy the game. Nobody will buy it. Nobody learned anything from it.
"The other day I traced a picture of the Mona Lisa and I did it just fine, I am now a renaissance painter"
Now, substitute yourself with my robot:
"The other day my robot traced a picture of the Mona Lisa and I did it just fine, my robot is now a renaissance painter"
This is such a broken line of thinking, I can't even begin to understand it.
In this particular case I didn't imply that I had done ANY of the work (or even that the result was any good - the "just fine" was a judgement on the conversion from Three.js to vanilla JS.)
Apparently the prompt was:
"You know, you should also build me a FPS. Insanely detailed quality pls, something that looks visually like it can keep up with AAA. Make it small in scope (like a tiny level) but otherwise impressive photorealistic graphics"
According to https://x.com/felixrieseberg/status/2119927483994907024
lies
edit to downvoters : Link in the post is not one shot anyways. It even says that on the page.
This is the one-shot version linked in the post https://claude.ai/artifact/2UXyB49DrVHFfhTMdNDFTA
What engine did it use to make this ? Unity ?
It's using Three.js, but everything else (physics, models, audio, procedural textures etc) is written without additional libraries.
If you're signed in to Claude you can duplicate a copy of the artifact and then ask questions about it or look at its source code in chat.
Here's a transcript where I did that: https://claude.ai/share/c0219852-fe26-48eb-8b24-bdf02823bed5
Fantastic. This is awesome.
Further to this, you can do this even faster than Claude, you can just download Unreal, they have ready to go demo games.
This is not as amazing as people seem to think.
No game ever was as simple an idea as "Make an FPS". The devil is in the details and that is where LLMs break down.
I still can't believe it's allowed to steal the world's content and sell it in a derived way like this, without going to prison.
"Steal"
I welcome you to read this - https://mises.org/library/book/against-intellectual-property
Which game did this steal content from exactly? Basically everything is derivative...
I'm not familiar enough with gaming to know where this crossed the line.
Rainbow Six? Call of Duty? Battlefield? Counterstrike?
This is exactly why this is so evil. Someone makes a porn using your face and your brother’s body does not make it any better.
What if SNL makes a skit by dressing up a comedian as you that kisses Ryan Gosling playing your brother’s body and airs it on national television?
This is a disingenuous argument and you know it.
I don't know it. Could you specify how this FPS is a derivative work while the mentioned games are not?
It is worst than steal. Since stolen content is at least what it is. This thing is stealing something and trying to say it is created.
Please, link what this game is derived from. Facts matter.
Are these models trained on the games. How are they generating such good graphics.
One perspective on the big AI corps is that they’re a grand experiment in capitalism: given a letter of marque to maximally exploit the combined intellectual capital of the world to date, what’s the most you can make? It is very unfortunate (although arguably also very predictable) that the answer is “mostly a bunch of hacks and ripoffs”.
[dead]
[dead]
Fine, so AI can make generic shooters. But the interesting thing about games is not graphics, but engaging mechanics. Like enemy AI, not in the sense of a chess engine destroying you, but with a theory of mind that guesses where you are and what you're doing, so you can manipulate and outsmart them.
Interesting games are the ones where you can hack your way into places where you "are not supposed to be," and games where you finish and think, "I could have done it better," or "What would have happened if I had done it differently?"
I don’t know but your response seems like a lot of cope.
I've been building an FPS since I realized Opus 4.8 claude had some essence of taste and modeling capability. I have 0 experience in game development. Along the way I learned how hard video game development really is and have a new respect for indie studios. AI will never replace original story telling and artwork - having a firm foundation and building the first models and level myself was key.
Since then, each model advancement has made it easier and easier to delegate and form workflows around each aspect of making my game - to the point that I login once and awhile to sculpt, get visual feedback for textures lighting and level design... I get a level done a week - test it myself, and ship it to beta testers who auto update and can send me anonymous feedback - all thanks to Opus.
I'm one dude - and I think this is a crippling vision to some of what's to come. AAA studios certainly know and will serve you AI generated games instead of paying talented people. I hope those talented people getting displaced can get enabled cause what I've done is pretty amazing.
AAA slop has been going on for years before ai era where slop become a word. It's not about mass production slops, it's about making something that genuinely sells.
Not saying you are producing slops. Just look at what truly sells on steam chart. The game industry is a creative industry just like others. Anyone can bang together some music or videos and upload them, but which ones are really getting popular and making money? That's the real question
When my all in stake is about, oh, say $5k - how can you afford not to try? Making it good is tough, I've been at this for nearly half of a year and I'm only now hitting my stride ( a level per week )
Especially if it's just gratifying to say you released something you're proud of. I have no illusion AAA titles with big budgets/advertising budgets and higher polish will give it a slim chance but,
Its better than a lotto ticket I'd wager.
Steam is full of AI generated games making $0. It takes a lot of work to make a good game even if AI does all the coding.
What are you talking about? Even charitable read of the current FPS scene shows biggest names rehashing the same stories and worrying more about presentation that anything halfway interesting. AAA is over? Good riddance. It is not over but AAA studios will just send slop?
Dude. They are already doing it. The only difference now is that you can too.
Hopefuly, I gave you an idea of what you need to do.
Sure, they've been sending more slop lately but at least there was some aspect of providing a livelihood for game devs. It sounded like that was the comment was expressing anxiety toward, at least how I read it. I guess game devs will just have to shift to something else, but as a (non-game) dev from the before times it's really not clear where to shift that energy. Especially if you used to depend on it to provide all your essentials for life.
I hear you and I don't really have a solution. However, what I suspect is that status quo can't go on as-is anyway. I can't say I am not worried about what is on the horizon, but just propping it all up so that company can release Modern Warfare MM seems wrong on a different level.
Anyone can asset flip an FPS in unreal engine.
I guess my question is whether it actually modeled the assets, or simply pulled a bunch of cc0 assets off the internet.
This has actually be a challenge for my agentic gamedev efforts. Getting to a particular visual style outside of PBR assets pulled off of packs online is more challenging right now.
It's all code. All math. Models, animations everything. That's what's impressive about this; not the game itself
This isn't a completely original approach: for example kkrieger did this some years ago as part of creating an extremely small FPS. It is not too surprising that it's something that is more suited towards LLMs (and vice-versa).
That is pretty impressive.
Gameplay is mid, but the graphics are pretty.
How much in credits did this cost?
https://x.com/felixrieseberg/status/2119925633564463609
> Everything in this game is from scratch: models, sounds, textures. ~$550 at API pricing, ~2.7M output and 445M input tokens, heavily cached. Took 8 hours!
I might just suck at gaming but this was miserable to play on mobile.
Yeah, I died fast.
Fighting a bunch of drones in super annoying. I had to move to the desktop to try it as the D-pad doesn't work on iOS.
https://nitter.click/felixrieseberg/status/21199256335644636...
Couldn't play it. It captured my mouse without centering it, so it was not possible to look around; viewport just kept spinning up or down.
Ai can't implement a cs style netcode, and that's the only thing I need right now
How is the discussion here so negative? I am frankly astonished that this was possible to generate, even with human guidance. No idea why everyone is in such a rush to dismiss it as a reskin, or complain that it doesn't work well in mobile. Most video games i have interacted with that look like this don't work in mobile at all???
Doesn't work on Firefox for me.
worked on Firefox on windows, takes a while to load though.
Works on firefox Android for me
It is impressive, i just don't see the point of it
I never spend more than one minutes on this kind of thing.
Don't crouch forward...
Maybe with another year of research we can train an AI model to design UI without adding 500 words of explanatory prose to every screen.
The more examples there are available for something, the less words you need to describe it as it can just infer the gaps and come up with something that seems OK.
Ran out of rifle ammo, died when my small arms fire couldn't hold off the drones. Realized I can't see them that well but they can all see and pew me pretty easily. Really says a lot about where we are heading.
note: hit a bug on bog standard chrome where I cannot toggle between "mission" and "holdout" once I click holdout. Is it a commentary on where we're heading too? Was this formally proven to follow a spec during design?
It does seem, across a range of fields, that the main benefit of AI is that it decreases the cost/time required to produce dreck: documents that no one reads, websites that no one visits, games that no one wants to play.
That's a lucrative technology, sure, but I'm yet to be convinced it's a useful one. We already had more lazy asset flips than we wanted.
I believe it drastically decreases the cost/time required to produce stuff. Some people are picky about the specifics/quality of produced stuff, some aren't, and some are welcoming specifically the quantity, for less than savory reasons.
It's a terrible game in the same way 10 years ago asset flips that used public tutorial blueprints and stolen assets from the unreal store were terrible games. We properly pointed out how big of a waste of time and how "unethical" for lack of a better word these games were, and we should do the same here.
A keen eye will also notice the immense amount of small random objects clipping into each other, the lack of culling, the at-first-glance models making sense unless you sit still to study them, such as the floating stairs directly above the player in the twitter clip mentioned in another comment.
Maybe it's incredible that a model can come up with something like this, but I find the spam of low quality slop with no direction or human input behind them to be kind of disrespectful to everybody's time. Who are these experiences for? Other than giving Anthropic $500 cause you're bored.
Maybe I'm biased because social game dev circles have been completely ruined by AI posters shitting everything up.
Asset flips were seen as unethical because people tried to charge money for them.
Nobody cared if you made a free demo using asset store stuff, that’s what it was for.
Cool! Derivative.
Nothing wrong with derivative but of course it's going to excel in well trodden areas.
So surely Anthropic could prove this is one-shot by just sharing the whole session right?
of course! not like the IPO depends on it, right?
Read both to these comments picturing Dario cringing and sweating.
I played it and it was fun, I absolutely felt suspense and performance was very good considering it's a web game. Music was fitting but still quite bland. The atmosphere looks good, but was obviously designed to show off advanced graphics...
I like how there were hiding spots and how the bullets of the enemies can reflect off the terrain, allowing you to find hideouts and gain height leverage to take out ground enemies. it's hard to determine the height of drones but could be intentional. I also like how the game certainly isn't too easy. Unfortunately the keybinds aren't rebindable, which is a shame because that means I can't use the visor on a touchpad. Also the undersides of some of the stacked shipping containers aren't properly rendered. Drones clip through some the containers in the distance, probably because they aren't there and part of the skybox, and drones interact weird when colliding with containers quickly jumping up and down to compensate.
7/10 please never generate something like this again.
The D button doesn't work so I could only strafe left. Good times.
Disable the vimimum browser plugin
I enjoyed the challenge.
I really don't get the religious fervor here. Artists can now focus on what is the actual ( and claimed by, sometimes, the same artists ) important thing: gameplay. Isn't that a great thing that we can get some of the crud out of the way? And if you want to do handcrafted engine where snow is generated as proper fractals, it is still an option.
But that is not the real objection is there? The bar has simply been raised.
Edit: FWIW, stories in games were getting away with murder so it is about time that changed.
Edit2: Geez man. Talk about apt game title too given the response.
High-quality games come from people caring deeply about every single aspect; I just don't believe there's any reason to think that someone phoning in most of it will put sufficient care into the aspects they do themselves, or even notice the lack of cohesion between elements.
If I made a game, I am confident it would be as bad as this one.
How did they get those graphics O_O
What engine is it using?
wooow. this would be easily several months of work 2 years ago.
you could easily put some idea behind, put donald t there, or something trendy now, and get at least some attention/fun, and maybe some money on ads . the capability is there.
Reminder:
https://fixupx.com/v01dpr1mr0s3/status/2119993017717358913
> Anthropic injects a hidden <ip_reminder> in Fable 5.2's turns via API;
> Fable gets obsessed over it and Anthropic blocks the request for "reverse engineering".
Wow, this is a terrible game lol.
Among games written for $550 in 8 hours including the art it's pretty good.
It's really not, you're just incapable of being honest with yourself on this.
$550 spent and what value was gained? No one was entertained, no one would buy it, no one will remember it by next week.
$550 wasn't spent. That was the "API pricing" value, but the author was on a monthly subscription plan so the tokens were already covered.
Just because they aren't charging that now doesn't mean their investors aren't expecting to charge that in future.
Monthly sub is a lie and should not be considered when looking at pricing.
The guy who posted this is literally a manager at Anthropic, so for him this was free of charge.
Oh, I missed that. Yeah, Anthropic employees all have effectively unlimited tokens.
The value was the joy in creating it. We shouldn’t pass judgement on other people for things like this. I don’t enjoy collecting coins or bird watching, but don’t mind if other people do. The amount of money spent creating it changes the conversation a bit, but people spend money on trivial things with no perceived value.
We’re moving into a new era where AI is used by a certain range of people (computer enthusiasts) like the earlier days of the internet. Wait until grandma starts using AI like when everyone started using the internet. Look at how that turned out. We’re not there yet, but that’s what you need to worry about.
> the amount of money spent creating it changes the conversation a bit
Both coin collectors and and bird watchers (and members of any other hobby) would look askance at someone who poured money into the hobby but didn't demonstrate passion otherwise.
Two common ideas in our culture are that [resource] is not a substitute for passion, and that buying something is not the same as creating/earning/deserving it. These aren't new ideas, and they don't suddenly disappear because AI is here.
> where AI is used by a certain range of people (computer enthusiasts)
I'm not sure that this is the case; I think AI is used by AI enthusiasts, and a lot of the pushback is because they're joining [thing] enthusiast communities without actually caring about [thing] or the community.
[dead]
No, it isn't. With $550 you could buy ten really good games made by someone who knows how to make games, or spend it on something else you enjoy or invest it. If he wasn't a manager at Anthropic this would be $550 of money wasted. Nobody will enjoy the game. Nobody will buy it. Nobody learned anything from it.
You just confused the cost of producing a game with the cost of buying a game, lol
It's not a game - its a tech demo.
We've already seen that it can make web games that run poorly and nobody wants to play, it isn't demoing anyrhing new.
I'd be interested to see the first one that makes >0$ on SteamSalesBench.
three js already has loads of tech demos, don't need Claude doing more for them.
for a one shot prompt and web based, it is not terrible i think. I mean no one is going to pay for this game
Does it forego using a game engine?
Like all these "games" it's using three.js
The other day I had Claude Opus 5.5 rewrite a three.js prototype - https://static.simonwillison.net/static/2026/Hidden-Reef-Egg... - in vanilla JavaScript and it did it just fine: https://static.simonwillison.net/static/2026/Hidden-Reef-Egg...
"The other day I traced a picture of the Mona Lisa and I did it just fine, I am now a renaissance painter"
Now, substitute yourself with my robot: "The other day my robot traced a picture of the Mona Lisa and I did it just fine, my robot is now a renaissance painter"
This is such a broken line of thinking, I can't even begin to understand it.
It seems a bit broken but you've added some steps that were not in the post you were responding to.
In this particular case I didn't imply that I had done ANY of the work (or even that the result was any good - the "just fine" was a judgement on the conversion from Three.js to vanilla JS.)
You appear to have a bee in your bonnet.
But just think how quickly you can now make so many terrible games.
[dead]