Most angles have nice consistency in this AI generated 3D room. Objects have pretty OK geometry. Some ghosting. But regardless the pace of these models is just incredible.

Prompt image here.

Looking great! I tried this the other day going through nano banana (2nd image). Results so-so: via results (not very optimized app: )

The progress is staggering. It feels like we're skipping entire steps in the 3D asset pipeline. What was once weeks of modeling and texturing is becoming a single prompt.

Yup

Is there a german word for frustratingly good? I’m almost irritated at how good this is.

Ahahaha. Totally.

is it generating everything besides the prompt image?

Yes

Is it generating a 3D model which is then viewed (traditional way) or generating the full 3D view on the fly (generative way?)

It generates a full 3D representation (splat + mesh) that can be used in whatever (three.js, unreal etc)

@BenjaminDEKR Hunyuan 3d ?

@BenjaminDEKR World labs

Consistency looks solid, love it.

damn, that's really impressive!

wow

This is really strong!!

Love this art style. Kinda like Borderlands. Generative game engines will be nuts

wow, @martin_casado what did you use to achieve this ? this seems perfect for something i’m about to start working on

Now turn it into splats and you have a game level

It is a splat.

3D gen is leveling up fast.

Amazing. People (even in industry, sadly) still think AI isn't a match for AAA title. What they don't get is AI is doing to music what DAW did. You don't wanna match AAA, you just wanna get out your idea fast, maximizing game feels and mechanics, without gray box placeholders. We're there.

is this tool available to the public?

the model hasn't been released yet, no. It's still under pretty heavy development.

This is impressive

impressive rendering!

Impressive work on this model

Ohhh this style is 💯🔥

Yeah works so well.

what engine/model?

World labs

Thanks! Going to look into it....

new lethal company map?

the speed of progress is wild. what's next, auto-generated gameplay?

Can you zoom in to see the details?

very impressive

Good geometry can mask overfitting. Test with novel object topology for robustness.

the public examples seem to have much narrower bounds - is this something actively being worked on? how were the bounds in your experimentations?

@michae1becker yeah model is still under development. What bounds are you referring to?

the world bounds. it's a very small 6DoF cube

@michae1becker Those images are from Dec. it's improved a lot since. Even back then it had multi-cube. Open spaces are harder, but for confined spaces including multi-room, complex architectures, and objects, it's quite good.

How does one get access to such a system

Looks great. How are you handling view‑dependent effects when mixing splats with a reconstructed mesh? Any eval on scale drift or normal consistency across a room‑sized loop?

What model is this

World labs

Use case: AI should be able to tell me where I have kept my car keys? We no longer need to remember things.

Which model is it?

How do I do this?

What did you use to create the 3d room? Any image to world model?

🙌🔥