Hacker News
points by arshxyz 2 days ago | 0 comments
Scientists say this is the first occurance of the word "model" in the title of a post on HN that does not refer to a neural network of some sort since 2023
[–] xnickb 2 days ago
You're laughing, but they're probably going to use these models to train a model and allow players design their characters from a prompt.
[–] superdisk 2 days ago
[–] giancarlostoro 2 days ago
Do they update this frequently enough?

Personally, I'd never build on top of Roblox itself, but... and I hate to say this, if they made a generic 3D service that let you export to any game engine, I think I'd pay Roblox for said tool if it was good enough, which, they have so many games and capital, I could see them pulling that off, I mean they already have a base model.

I would love to see more text to 3D model models that can wire up straight to a game engine, I think we'll see a lot of interesting things, like children prompting the worlds of their imagination into reality, and the 50 million memes that will plague social media.

[–] embedding-shape 2 days ago
> I would love to see more text to 3D model models that can wire up straight to a game engine

Same, but none of the models actually have good enough output yet to make that work. Currently they're all aiming to "fake" the niceness by doing modeling+texturing, but when you look at the wireframe, it's pretty clear they're very far of from being usable, even less so in a game where performance matters a great deal, and you can't just YOLO assets however you want, which is what the "text to 3D" models like to do.

Inevitable that we'll get there eventually though, it's miles ahead where it was just 1-2 years ago. I guess my point is that it's not the "3D model > game engine" that is lacking at the moment, but "text > 3D model" that cannot produce high quality enough results.

[–] giancarlostoro 2 days ago
I wonder if the solution is either an intermediary "model spec" that can render into others (this becomes the target for all these model models), or an LLM style hack where its trained on recreating 3D models using code, I know for CAD (I forget which one it is) there's a CAD program where your models / designs can be written in code.

One weird thing I run into is, I can have Claude make some interesting things with threejs, but translating it to other engines isn't always so great, and its usually raw code. I'm not a 3D guy, I just want to get up and running with things I can code / think of.

[–] whh 2 days ago
Design models with a model trained on models. It’s quite straightforward
[–] Buttons840 2 days ago
All models are wrong, but some models are trained on models of models and still wrong.

https://en.wikipedia.org/wiki/All_models_are_wrong

[–] xnickb 2 days ago
I'm sure one could recreate a look of some popular model. So the model trained on models then would generate a model of a model.
[–] GaunterODimm 2 days ago
but I'd prefer to have an industry expert involved, for the accuracy of the model, model to supervise and co-create model trained on models then would generate a model of a model.
[–] mechanicalpulse 2 days ago
Model the model model on the model model model and you'll be fine.
[–] icepush 2 days ago
As long as the models are not modal, this will work. Modal models are a miserable method of momputer minterface mesign.
[–] talon8635 2 days ago
I modeled a model that’l allow you to model your own model dawg

How’s that for a throw back

[–] dylan604 2 days ago
As photographer, I find it hard to work with your model. It's like it has no training and just sits there waiting to be told what to do. I spend more time explaining what I want the model to do than actually working with the model. Maybe I should shell out for a professional model instead of working the the open models hoping to be a professional one day.
[–] asciii 2 days ago
> Some examples include models of Eve from Stellar Blade, Mei from Overwatch, and a Horse cock dildo.

Lmao!

[–] WJW 2 days ago
For "research" purposes, no doubt.
[–] bilbo0s 2 days ago
You laugh, but this is the kind of data that allows you to train models that actually catch a lot of corner cases, as well as generalizing to other 3D applications.

I'm not gonna comment on the morality of mass acquisition of all of this data, but as someone who deals in 3d in the medical/dental space everyday..

there's absolutely no doubt that all of that data is as valuable as gold in research.

On a tangential but related note..

I mean, people in possession of troves of data, especially in the medical or 3D industries, really ought to start understanding the value of the data they're sitting on. You can give it away if you like, but don't be upset when someone turns it into a couple hundred billion dollars and doesn't compensate you.

If you won't like it if tech companies rip you off, then take the legally required steps to protect your data. (That includes making sure you licensed your data properly. And protecting your data with explicit use licenses and, potentially, even access controls.)