Neat! But not quite what I was expecting. It doesn't really make a 3D "scene". It makes a 3D panorama (building a mesh based off the depth of the panorama). Move off the center and all the things that were occluded are missing.

I guess, because we're in the time of generative content, that I was expecting it would recognize what's in the scene and fill in the missing parts. Maybe in version 2

I really hope there is like an "AI" flair on projects like these, i love what they aim to do but man the saturation levels mixed with "enough to make it work in theory" effort makes me want to filter out these posts.

I understand its all vibes and fun, but like at least some warnings that "this thing might lag your computer" and buttons that work.

We really need models that produce intermediate files. A .blend, a whole godot project, an actual krita/psd file.

Things like this product such cool things....all the way up until you need to move a specific tree half a meter to the left, and then you're either stuck in some prompt/seed hell or have to open an abomination of a file in an editor.

An opensource Blockade Labs Skybox.ai sounds good and thats a lot of tooling in your pipeline so congratulations putting it all together I would 100% clone this on a HuggingFace space !!

The splat outputs remind me of Cyberpunk 2077's braindances

Optimized the webpage layout for mobile users.

Not really proper 3D, more like a kind of 2.5D.

Looks like only nvidia cards. Can anyone guess how much vram you need for this?