▲ 277 points
back
88 comments
In case it's not obvious, this isn't a model, this is python scripts that call out to models (code not available).
It's mostly "did you know you can attach an LLM to a PCG system?", but there is one idea here you don't see much of: an image model performs the composition (which image models are really good at), and then you extract the objects into 3d via things like SAM3D before placing them in the world, which is pretty interesting.
The rest is standard stuff you'll find in your favorite PCG system/game engine. Still, think most people don't realize how good LLMs are at 3D these days, especially Fable 5 (which this project predates).
LLMs are a PCG system. This is one of my pet peeves, I've been working on a procedural generation constraint system for quite a while (way before LLMs) as a hobby, that came with a domain specific language, where you could specify constraints.
Things like the red key must be reachable before the red door, and there should be no path that bypasses the red door, etc. written as a set of constraints.
I was working on is because I realized, constraints tend to be similar in nature (graph cut, minimum/maximum shortest path etc.) algorithmically, but its usually quite hard to control designer intent.
The reason I shelved the project, was I asked an LLM a couple years ago, to generate a dungeon room graph, following a similar set of constraints, but stated in plain English, and it did a perfect job.
Which brings me to my final point. LLMs are amazing and flexible constraint solvers, and are woefully underutilized in scenarios like this.
The trouble is the LLMs non-determinism and the "You're absolutely right, I placed the red key behind the red door rather than in front of it. This was load bearing for the whole dungeon and I blew it" moments.
With a hand authored constraint system if the red key is in the wrong place you have a bug in the constraint system that can be deterministically fixed, with a LLM you have to roll the dice on prompting again and go through expensive verification.
> Still, think most people don't realize how good LLMs are at 3D these days, especially Fable 5
Can you explain more? Can I use Fable 5 to generate a 3d model from a couple of photographs?
> most people don't realize how good LLMs are at 3D these days
It seems to work well generating static 3D models. However typically with way too many triangles and no skeleton unless it is a human.
That's my experience anyway. I would be happy to be proven wrong.
It looks impressive. However open worlds are at their best with hand placed details and environmental storytelling. Look at Skyrim/Cyberpunk vs. Starfield where most of the world is proc gen’d.
I get that you run this, and then edit it. But the generated villages just aren’t interesting in my opinion. It’s probably great for tencent’s market where you’re mass producing gacha style games, but I don’t think worlds made in this fashion will scratch the open world itch like the best open world games out there.
You know what's also good at storytelling? AI. Or else we wouldn't have writing awards being revoked when it comes out it was written by AI.
I wouldn't say we don't need artists or writers, but I think most people will be surprised how quickly fully AI driven pipelines will advance here.
I'd actually be curious to read an award winning ai book, all the ai books/interactive story generators I've tried have been... Ok at best... Do you know what novels have been written by ai and award winning?
Its a bit like saying AI is good at math when its being used by Terrance Tao. Like okay if its so good at storytelling, use it to generate a good one and share it under this comment for feedback and we can check out how good and original it is.
> You know what's also good at storytelling? AI
Writing a coherent story =/= good at storytelling
This is a great use of our global water supply
What award/book was this?
Yeah, all that hand place details and humans do it better crap only last for like three months based on the recent past.
It's at the very least very good starting point, devs should still go in and add mojo to the output
you can do both, generate a huge world and orovide aritsts with aithoring and collaboration tools to dress it. to prevent starfieldlike stuff -__-
In both the Autumn and Winter examples in the hero images, the algo seemed to have placed buildings on the water in the foreground
In the summer example in the hero images, the building placement + small pockets of water on the left looks odd and low attention to detail. A similar poor quality result as if an uncaring human used a scatter brush...
Curious if the examples are cherry-picked and by how much, or if this is one-shoted
I am sure they are cherry picked. No one would spend time trying to publish results like this without picking the best results they possibly could to showcase.
In this competitive world, you need to integrate the cold plunge deeply with your lifestyle to stay ahead.
This kind of content generation can be quite cool to enable new game ideas for indie developers that were only possible with AAA before. But it also makes it hard to gauge the amount of human work that went into the game. When you see an elegantly designed building, environment or easter egg in the game, was that really made by a human (or even prompted by a human), or did the AI generate it autonomously as part of a much larger generation step?
> When you see an elegantly designed building ...was that really made by a human or did the AI generate it as part of a much larger generation step?
Does it matter? Why?
A good painting is never “just a good painting.” The context, the artist, etc. are part of the art. To remove that is to make it no longer art because it’s no longer an actual expression of anything personal. Pretty words with no intent behind them don’t mean anything.
Plenty of Holiday Inn’s have lovely painted landscapes on the walls. People aren’t paying for them to be in their homes.
There are millions of games. When choosing which one to play, you've normally chosen one that looks polished over one that does not, because of what it signals. Now that signal is gone, can't easily discern what's a sloppy crappy game vs one that's actually fun to play.
To me what matters is being able to verify human authorship at a glance. It used to be that when you saw a work of visual art you were automatically 100% sure that someone had to have spent a great deal of effort producing that, but image models took it away. This is the same situation but with game environments.
Interesting prompt to world pipeline. I wonder if there can be more style applied than just the "MMO cartoony" style? Like medieval Witcher or realistic Cyberpunk buildings?
I'm sure the answer is yes, even now there are pretty good model generators out there that can do realistic models. Experimental things like this just tend to go with cartoony styles since mistakes and lower quality (read: faster generation) can be chalked up to "lol", whereas realistic styles tend not to be forgiving with art mismatches and shading mistakes.
In modern polished games, the art team spends a lot of time tweaking things to make sure it looks great and coherent. Most people are not willing to have their agent run for a month to do this work. I don't know if it's been tried yet but I would be really interested to see the result.
Again, that Claude web design aesthetic is just seemingly everywhere right now. It takes me out of what I'm reading as soon as you see the tells.
The water levels don't change seasonally..So I'm guessing it's a very high rainfall area? In which case erosion and nutrient leaching are bound to be issues for the agentic farmers. I wonder if they've thought to build soil with perennial ground cover species..
But as I say this, I realise the tree species seem to change very dynamically from season to season. So I'll have to assume either the temporal rate of change is actually a lot slower than one would expect in a model like this (allowing large tree species to regenerate annually), or there are new coniferous flowering tree varieties in this universe.
I just left the website open by mistake and went to take a walk. The temp got 80 degree celcius. My laptop's chasis got really hot
I kind of feel like this is how Crimson Desert was made.
It's enormous, but utterly devoid of any soul. Like eating a $3000 subway sandwich.
Unfortunately, code is not available.
Wow, the methodology in the paper is so detailed—I'll try following it!
The assets look good, but for some reason the terrain looks too cartoonish to me.
Looking forward to hearing more updates from tencent 3D models. :)
I hate to be an HN bikeshedder but why is everything "Claw" recently?
Because it's a buzzword. It causes buzz. Putting it together with "world" makes it buzz even more because people want to talk about world models all the time (it doesn't matter what the actual input and output is, just that it's a world model, not some meagerly language model, it models all the world, much better than modeling just boring old language), as well as about claws. So that's why.
Branding is amazing. In this case with a normal word, it's a sort of network effect?
Peter Steinberger, creator of OpenClaw, just spoke at YC Startup School. The entire thing is worth a view.
> There are 33,000 "claw" named repositories.
It's just the natural evolution of AI into a crab.
impressive, but as I understand the worlds were pre generated an there no way to reproduce it on my setup?
It looks cool. Shame the source isn't available.
But Claude can do this with little prompting.
[edit] Not sure why the downvote, but Claude Design can do this, its just a matter of knowing the exact prompt.
This was created with just the prompt on from their front page and nothing else.
That's literally a cut+paste. If I had spent time on it, you can get it close to what's shown.
Claude also created that as a Blender file with assets for each of the generated items.
This looks so cool. Nice work :D
is this dwarf fortress
Looks impressive, haven't looked at the code yet, I'll at night.
There is no code. The GitHub repo is empty.
What code, isn’t this just LLM output