I have experimented with this type of modeling for a RTS game I'm building: https://playwarpact.com
Few things:
1. Opus 5 is just as good for this tbh and cheaper.
2. They don't generate optimized 3d models. They have high poly count for simple geometries.
3. A better approach I've utilized for game ready assets is to use the model to create low poly silhouettes of the 3d model and then bake textures that include a lot more details like windows, doors etc with tools like Meshy. Can post a tutorial if there's interest.
The models need some more RL to be able to do this autonomously.
> Opus 5 is just as good for this tbh and cheaper.
It depends on the fidelity, there's a threshold slightly over "low poly" that Opus cannot get over. Once you get into creating foliage l-systems, or physical mob animation, or a house with realistic appliances, Opus is under the convergence threshold for a world model no matter how much time you give it, it will flail around and say it's done even though it's nowhere close to plausible. Fable takes forever but knows how to zoom in and out on the abstractions.
> They don't generate optimized 3d models. They have high poly count for simple geometries.
By default LLMs will do the quick prototype thing and slap some primitives into a THREE.js scene. Which is probably correct because most users don't know or care. But if prompted Fable will connect the manifolds/sculpt with a mesher, instance out the pieces, etc. and make you an efficient art pipeline. You can make it optimized. Just not in 10 minutes.
> A better approach I've utilized for game ready assets is to use the model to create low poly silhouettes
You can also let Meshy make the geo and then fit it back to the world with PCA + BB's. Diffusion models are still 10x better at making physically plausible geometries than LLMs, plus they are faster all things considered. One technique I've found that works well is socketing; let the LLM generate the high level structure with blockouts for sockets, then slot in the higher poly objects that fit. Which is closer to the professional approach for world design.
I’m interested! I struggled to get goos detail on my Back to the Future Delorean 3D animation on my docs hero: https://contextify.sh/docs/
I posted this previously and people wanted more info so here is how I built it, with a deeper dive into some film scene recreation I worked on focusing on a scene from Apocalypto:
why would you force me to login to see what the game is, I have zero information and you're asking me for a password or my google information. hostile and creepy.
These things always look impressive but I fail to see how they are usable beyond demos or simple games. In my experience the topology is often a mess and texturing is difficult as well
i discovered doing this in claude design and i can tell you... it's useful in a professional domain. Although I am not going to share what I'm using it for, to keep an edge in my domain ;)
I wish you would provide more information. About time, costs, reliability (how many duds did you have? did you need to do any babysitting or etc) and so on
But it is a very impressive looking demo, for sure.
this is a one-shot result but i have a really really lengthy prompt: https://github.com/PhiloLabs/fable51-worlds/blob/main/union-... with clear guidance in using subagents and self-QA loop.
~2 hour (extensive subagents usage), total ~8M tokens, ~$33 under API
yes experimenting with it actually, will update here! in fact we've generated most of the tourist spots in sf, should be reflected in the repo soon too
This reminds me of the experience VRML wanted to be 3 decades ago. The only difference is that links went to other worlds living on another server. But that seems trivial to implement with this.
Cool demo! That's funny. I literally sketched out something similar 30 minutes ago on paper. I wonder if there was some recent, semi-related article/link that subconsciously prompted the relevance.
we're working on a survey paper about world modeling via code, with folks from frontier labs (qwen omni, oai, etc) and academic institutions (e.g., oxford, stanford, etc) reach out to us about collab: team@philolabs.ai
this is really neat. I wonder if people will start building open world games or "AR" games in the vein of Pokemon Go and Ingress based on similar tech. I've had a similar idea for a long time but don't think it was feasible before now due to AI
yes we're working on it! trying to push a few open world rpg games with real economy and game design. it's also super interesting to benchmark the current models' capability in this direction, since this is a naturally hard and multimodal coding task
Cool. But the question remains, what is the point of this? I don't mean to pick on this project specifically, but it's indicative of AI usage in general. You end up with something that's like "wow an AI made this" but it's not really anything that interesting or useful in and of itself. I've yet to see a single agent generated piece of software above that bar.
the topology/texturing critique is fair for mesh generation, but code-generated worlds mostly sidestep it. when the model writes three.js or blender scripts, geometry comes from primitives, csg, and parametric construction, so topology is clean by construction rather than something you clean up after. texturing is still a gap, agreed, though procedural materials cover more than people expect.
Because it’s a very complex task to do especially with any level of quality and consistency.
Classical algorithmic approaches for data to CAD are quite unreliable heck commercial CAD software still struggles turning scanned 2D plans into sketches with any degree of reliability.
Something of this scale would normally require 1000’s and 1000’s of man hours in manual modeling, placement and quality control.
Some of the latest prompt to CAD agents and models I’ve seen are the most impressive use of AI I’ve seen in a long time.
At least for me this is rather exiting since it gives me the ability to turn many more ideas into reality as a hobbyist rather than spending the entire weekend in Fusion 360.
Why don't you try to do it yourself with Fable and compare it with this one? Even with AI there are still a lot of work, knowledge, and gotchas involved. What's the point of such comment?
fable 5.1 generated an interactive 3D union square, and the agent filmed its own tour guide vid inside it. you can walk Powell to Stockton, read the actual storefronts, cross a working intersection, watch a cable car go by. went inside Apple and the Nintendo store, lower level included
Few things:
1. Opus 5 is just as good for this tbh and cheaper.
2. They don't generate optimized 3d models. They have high poly count for simple geometries.
3. A better approach I've utilized for game ready assets is to use the model to create low poly silhouettes of the 3d model and then bake textures that include a lot more details like windows, doors etc with tools like Meshy. Can post a tutorial if there's interest.
The models need some more RL to be able to do this autonomously.
It depends on the fidelity, there's a threshold slightly over "low poly" that Opus cannot get over. Once you get into creating foliage l-systems, or physical mob animation, or a house with realistic appliances, Opus is under the convergence threshold for a world model no matter how much time you give it, it will flail around and say it's done even though it's nowhere close to plausible. Fable takes forever but knows how to zoom in and out on the abstractions.
> They don't generate optimized 3d models. They have high poly count for simple geometries.
By default LLMs will do the quick prototype thing and slap some primitives into a THREE.js scene. Which is probably correct because most users don't know or care. But if prompted Fable will connect the manifolds/sculpt with a mesher, instance out the pieces, etc. and make you an efficient art pipeline. You can make it optimized. Just not in 10 minutes.
> A better approach I've utilized for game ready assets is to use the model to create low poly silhouettes
You can also let Meshy make the geo and then fit it back to the world with PCA + BB's. Diffusion models are still 10x better at making physically plausible geometries than LLMs, plus they are faster all things considered. One technique I've found that works well is socketing; let the LLM generate the high level structure with blockouts for sockets, then slot in the higher poly objects that fit. Which is closer to the professional approach for world design.
I posted this previously and people wanted more info so here is how I built it, with a deeper dive into some film scene recreation I worked on focusing on a scene from Apocalypto:
https://banagale.com/cinematic-canvas-ai-film-animation.htm
I’d love to improve the look of these things.
There is.
Fake edit: there is a longer video here 1 min long: https://github.com/PhiloLabs/fable51-worlds/blob/main/union-...
I would be especially curious to see the NPC person/car logic and if they're on rails or what, that's a pretty good NPC density for a demo.
Here ya go: - https://github.com/PhiloLabs/fable51-worlds/blob/main/union-... - https://github.com/PhiloLabs/fable51-worlds/blob/main/union-...
https://news.ycombinator.com/item?id=49525160
Classical algorithmic approaches for data to CAD are quite unreliable heck commercial CAD software still struggles turning scanned 2D plans into sketches with any degree of reliability.
Something of this scale would normally require 1000’s and 1000’s of man hours in manual modeling, placement and quality control.
Some of the latest prompt to CAD agents and models I’ve seen are the most impressive use of AI I’ve seen in a long time.
At least for me this is rather exiting since it gives me the ability to turn many more ideas into reality as a hobbyist rather than spending the entire weekend in Fusion 360.
Begs the question, an earnest one: what software would be above that bar?
check source code + more worlds soon
> Geometry is derived from OpenStreetMap (ODbL) and USGS 3DEP (public domain).
This is probably how I would have tackled this, saves on time, especially if Fable just writes code to convert OSM data to a reasonable ThreeJS model.
~2 hour (extensive subagents usage), total ~8M tokens, ~$33 under API