alt Hacker NewsFable 5.2 one-shotted my nephew's dream game
Comments
I feel for kids growing up with LLMs like this. They will never impress a parent, teacher or peer with a simple “hello world” type text game, when the peer could generate a full multiplayer video game with a couple of prompts.
The world has changed and we can never go back.
"one-shotting" is fun for a low (human) effort toy, but it's a mistake to get overly hung up on LLMs getting everything right on the first try, since a lot of that is the harness and the internal prompting that makes it burn extra tokens to test everything to get it in that one "shot" (ui turn).
While it's nice that LLMs are getting better at arriving at the desired deliverable with fewer surprises (or broken things) doing it is not necessarily a sign that the model is actually smarter, just an indication that what you are doing is likely a well-trodden path for the model, the RL and the harness.
My dad asked recently if AI is good enough to create something like DX Ball - a breakout style game we used to play a lot.
Opus 5.5: https://claude.ai/artifact/UuT7P936dU1XoPByYZ7xyD Fable 5.2: https://claude.ai/artifact/UszKFvuonWAT5psjMttAEg
There are some areas in which Fable is clearly better but subjectively the Opus 5.5 version feels more fun.
Both are one shots, Opus 5.5 was done a few weeks ago before they added "guardrails" around recreating games. Fable 5.2 changed some game mechanics despite my prompt asking for them to be the same as original game.
340M tokens is like $5k?
Edit: more like $300, because it was mostly cache reads:
https://news.ycombinator.com/item?id=50434388
I haven't used Claude in a while, what does the billing for Fable look like these days?
Last I heard they were billing it per token, even on the subscriptions? Or did they change it again?
Edit 2: Fable seems to be included in Max and Ultra but not Pro (aside from occasional promotions)
https://support.claude.com/en/articles/15424964-claude-fable...
It’s impressive, but it’s not really his. All the details aren’t quite right, some of the designs Claude altered (e.g. guy holding one knife instead of two). It’s good enough for him, but if I asked Claude to make something based on my idea, it wouldn’t be good enough for me.
I have experienced this myself, although the novelty kind of wears off after a while. It is still fun to make small games to co-op on the TV.
Yes, hard to believe but most of these games can be almost one-shotted in my experience, with bugs but still playable. We are not talking about some properly made games but imagine for a second that there are some number of casual players out there that wont mind spending 30 minutes on a game for just a bit of fun and having these just materialised just in time is a lot more convenient then hunting for them on the app stores.
For anyone interested, I put all of of my work into an open source tool that uses your iPhone as a console which then can cast on your TV. Up-to 4 local players supported. It can even run proper games via WASM. It is Apple Silicon only.
What does one-shot mean in this context? Aiden (the main character)'s walk cycle is kind of a mess and its also the animation that a player is going to see all the time.
It appears to be eight frames, but after taking a closer look, I found that they don't seem to be consistent [1]. Using ControlNets and various other strategies, you can definitely build out a "nominally acceptable" walk cycle [2], but you still have to do some manual work.
[1] - https://aidens-revenge.pages.dev/assets/characters/aiden/ani...
I used my $200 free fable 5.1 credits to build a WhatsApp sticker maker, it created a version I was not happy with, but it burned the 200 in one sitting. Node.js app, vue frontend, etc.
I didn't used the project a lot since I kind of hate it on how much was spend.
Last week I used just a little bit of my $20 pro account to build a replacement with rust end to end, just to gain some experience with the rust hype. Very happy with the result, not perfect but snappier, and even better: a single executable, I no longer need to fire up a backend and front service separately.
Ignoring all the positives & negatives surrounding this technology for a minute, this is an amazing outcome for the kids.
I think it's a super cool thing to play around that I definitely wish I had as a kid. But I kind of miss that magic that came from playing old retro games in the 90s, where the game would give you the scaffolding of a magic imaginary world, and a few pixels were enough for your imagination to fill the remaining part. It's crazy how immersive that felt.
So in that sense, this seems to go in the opposite direction.
I appreciate the token breakdown, but how much would this cost in money? Chatgpt estimated $400 to $600, but I don't have enough knowledge of token pricing to know if that's accurate at all.
This is both cute but also a bit sad in terms of game stylization, I guess these generic AI designs could become so normal to the youth that in the next 10-20 years it just appears everywhere
Year is 4098.
You prompt: " Sea facing villa with walls having 90% glass ratio, cozy basement, pool on ground and upper level. Delegate to a sub agent and when done, build a supersonic jet on my rooftop to fly me there." and all that molecular nano assembly happens in 19 hours.
In that kind of abundance - poor I don't know but would anyone be any richer? Would anything mean anything?
I quickly dove into the game and then got lost as I couldn't exit and had to spam keyboard to see the controls. The menu refers to "Punch" but it wasn't obvious what key that is (and it wasn't "p"!).
I see that ENTER pauses, but may I suggest also making ESC and/or P pause. And then the pause screen showing or linking the keyboard controls and an exit path.
(Caveat: I'm not much of a gamer so maybe the current controls are standard? Anyway, I feel like it could be more inviting to casuals with something like the above.)
What I wonder reading that would be how much more effort to actually port it to the PlayStation, or whatever game console. I recently asked my local AI to make an ebook reader for my Gameboy pocket ;) I just think about all the cross collaboration of devices for stuff we can only imagine.
> I must admit I never tested this with Opus 5.5, so I'm not sure whether this was possible before Fable 5.2 rolled around.
Since somewhere around Opus 4+ it's been really good enough to do these things, but 4.6 was the most pivotal version since a 1 million token context window can allow a model to fit so much "in its head" basically.
It's still some horrendous slop. The distance to some Mario Wonder is immense. Yes, maybe in a year it won't be so, but this distance, perceivably, is not contracting at all. Like, everyone sees that there is no decline in this feeling of "ai slop". You could do something with AI a year ago, and it was "AI slop". Now it's much more advanced and technically impressive, but still, "AI slop".
Kids are now coming of age with the software equivalent of a Star Trek replicator. LLMs are going to radically change the cadence and workflow for the production of computer games, no doubt. When you can turn out a professional product based on little more than doodles and some text prompts, it rather devalues the historical effort of bringing a unique game to the marketplace.
I rather hope that the novelty of generating this kind of toy wears off.
This is awesome, but I don't get the repeated use of the term "one-shot". There were many human-driven requests and iterations, and this is illustrated in the blog post.
This really was more like a normal person had access to a game developer team (with artists and musicians) and had a custom game built based on interviews and playtesting.
The section "The drawings and the game" is amazing.
When I was about 5, I saw this machine where art would go in, and much more detailed art would come out.
I must have been confused about how it worked (I think it was just a badge laminator and I had been confused about the inputs and outputs)... but it occurs to me that we actually have that now.
The magical machine my mistaken mind dreamed up as a kid is actually real now, and that's nuts.
I noticed this recently with Opus 5.5: you can tell it to "make this more fun" for a game project and it... makes it more fun.
It's neat (if unsurprising) that Fable 5.2 has this ability as well.
My loose testing shows other models aren't quite as good at this yet. Here's the prompt "Build a mobile-friendly three.js game where a mouse has to hide from a cat" by Opus 5.5 Extra - https://claude.ai/artifact/RRyEndXLSPQbcig3BgzgCZ - and by GPT-6 Extra High: https://squeak-and-seek-mouse-1027.simonw.chatgpt.site/ - the Claude one is a better game, even though they look very similar at first glance.
Update: took it about 40 minutes but here's Fable 5.2 Extra's attempt at the same prompt: https://claude.ai/artifact/Agsqbng2MnkTPAU67it87o
I wrote about Opus 5.5's sense of fun in more detail here: https://simonwillison.net/2026/Oct/28/vibe-game-development/
My intro to game dev as an 10 year old was learning to use my cousins C64, writing some BASIC text based games. It was pure joy discovering programming that way.
It is utterly depressing to think that kids will never take that path again. Instead, we will get pulling the slot machine lever, zero skill, zero effort, nothing gained.
"Creating something" now means typing a few text sentences in to some prompt and paying money to these companies that want to essentially make humanity redundant. Instant gratification, no hard earned skills at the end. Compare this to the kid building something themselves with scratch, improving on it, moving on to Godot etc.
The AI boosters in here saying how amazing this is are being incredibly short sighted.
Anyone doing this with their kids: please take a moment to think about the implications of what you are doing. Even explain to your kids what an LLM is, the moral implications (massivley capital funded companies have stolen so much content to make this possible and now want to sell it back to us) and the importance of effort and learning real skills.
This is phenomenal. Great work, dude. Man the fun my brother and I would have had in the old days with this stuff. It looks absolutely fantastic.
I no longer read or watch accounts of these models one-shotting things (unless you tell me it one-shot a differential AES cryptanalysis, or maybe an entire healthcare system with billing and radiology viewing, compression and annotation)
All these frontier models are capable of creating these amazing things. It's just reality. If it's in one-shot, it just means there's been more RL in that area, more reasoning is being spent in certain places, but overall you can get the same result without spending $300 in tokens if you feel like waiting a week or two for your claude subscription to reset.
It is very impressive though, and the implications on what a smaller version of this model can do is very exciting (aka what I expect Opus 5.6 or 6.0 will be)
I played for about 10 minutes and I swear - this was fun. There are a bunch of bugs to polish: collision detection is sometimes weird, you can catch enemies at the end of the map and just beat them up forever, adding tons of score, Pudding's animations feel quite off, and Stage 3 is VERY hard. However, the game still looks and feels quite polished. I loved the music as well. FWIW, I'm glad I'm not a game developer these days.
I have to say that this is amazing, a while ago I have tried making web games through AI but the capabilities were enough for simple games but failed for multi-level. Getting something like this with all images+audio is something I wasn't aware that could be doable in one shot.
I definitely need to try this.
It’s a lot of fun to build games. I started building little pieces of one[0], then a builder for it, then a story, then a video recorder. And there’s a lot more iterations in the roadmap. It’s great as a creative outlet
It’s not one shotted anymore or course. But in the process, it’s also cool one-shotting new features or experimenting with variations of things very quickly
Eventually it becomes similar to any longer lasting project. Complexity starts creeping in, maybe technical debt, and managing everything to keep it consistent/coherent slowly starts taking most of the time. And I’m not maintaining it for real active users, which would also become some other type of project
I had ChatGPT write a webapp to teach my kids letters. It has profiles, fun interactive games and keeps track of progress in a database.
It crazy how fast you can iterate if the building blocks (db, k8s cluster, authentik) are already present and you know what you want from the model.
Without reading any of your blog, I went immediately to play the game and my very first though was, "Woah, this is just like the ninja turtles arcade game."
Then I read, the prompt; well done!
Some parts of this seemed cool, and others made me terribly sad. I am not sure how to express it beyond - I really think inserting hulk as a character was a mistake as you don't have any chance of having intellectual property rights to the character.
On the other hand, it's very much like you photocopied some comic books, cut them out, and did a little home theater thing with them (only with even less effort on your part).
I tried the game and in some seconds noticed I can kick the enemies forever if I put them close to the left corner. I'm feeling now that I'm the AI exploiting bugs.
This is really cool. This would've been a dream for me when I was young.
One bit of critique of the Fable generated app here is that even though it technically runs on a mobile browser (Brave) it doesn't seem to be actually playable on it beyond playing the intro music.
> Fable 5.2 one-shotted my nephew
:O
> 's dream game
:)
I’m waiting until I can give AI a link to a Macintosh garden game and ask it to be reimplemented in the web browser.
Total cost was around $4k USD?
That’s a lot of fun. I’ve been also designing my own game. Testing fable right now
I understand you, I tried to give a huge tasks and getting fun with one shot result. But actually more interesting how long this process worked and how many tokens spent, because in GPT you can simply add target and GPT will work even 1 year non stop until reach this target. But I got what you talking about, but still time and amount of tokens is more informative information.
Just a small suggestion on the prompt. I usually also add something like "Bring in a critical Art Director, Animator, UX Expert, Staff Engineer. Don't release the game until they all agree it's AAA quality." I wonder if that would have helped with some of the issues/bugs others called out. I do this with Codex and it's been working really well and not using many tokens.
speaking in a boring party-killer voice:
That's technically impressive, sure, and if the kid loved it I feel happy for him, but...
I personally got bored after a minute of playing that, because that felt like a classic game with a slick skin over it. And games that really capture attention usually feel novel, not "Super Mario but make it about Dario".
Currently "oh wow I'm in the game" has the novelty value, but I feel that in two years that would feel like telling Nano Banana "make me look like a princess". Everyone can do that and no one quite cares anymore.
But the bright side is, with that capabilities level, people who have novel and unusual game ideas can probably implement them much quicker and easier, too.
Sounds like your children will soon learn hard work does NOT pay off
Is Pudding the dog (unlocked after beating the first boss) supposed to shrink anytime it jumps? I guess even Claude-turned-world-class-game-developer isn't immune to bugs. ;)
Honestly, setting all the oft-cited AI caveats aside, the end result is cute and perfectly serviceable as a game for the kids. I can't imagine how stoked they were about playing it.
Ooooof the results here are rough. A very young child might find this charming, but I think it's kind of embarrassing. Like, the walk cycles alone are. Yeah...
To say nothing of ur word shaders, inconsistency in the characters, questionable gender stereotypes, etc.
This definitely tells me that AI definitely isn't there yet.
It's not as good as Street Hassle on the Commodore 64. AGI.
> Make it one of the best games you can possibly make. You are a truly world-class game developer. Do whatever it takes to finish it and make it a best-selling game.
Honestly, I should use this prompt on my human self!
Yup, classic AI game! controls dont work, the character got stuck in one corner of the screen while smiling like an idiot.
Growing up, games were such a scarce commodity that you'd really suck all the marrow out and spend hours beating your best times, looking for Easter eggs, or messing with custom maps and mods. There was a whole genre of gameplay around just exploiting physics bugs to jump a bit further.
The other night I carelessly made an infinite runner game while watching TV and the feeling of sheer abundance quickly faded into a bit of depression. Without scarcity, nothing is really valuable.