World Labs: a 3D world from one image
On 2 December 2024 World Labs first showed a result: a system that generates from a single image a three-dimensional world to walk through in the browser; it is an “early preview”, available as a demo and a waitlist.
Why it matters
Generation moved from pixels to a scene one can move in, whose look does not change behind one’s back. An editorial assessment: this is a demonstration, not a measured model; the demo was public, and an outlet opened it and described its limits.
By the post, once generated the scene stays fixed, is controllable in real time and has “correct geometry”: worlds “obey basic physical rules of 3D geometry”. The camera can be varied (depth of field, dolly zoom), effects can be applied to the scene; text enters through a separate image generation by FLUX, Midjourney, Ideogram or DALL-E. A few artists had early access. What the record does not state. The post names no model, size, data or measure of consistency. TechCrunch the same day, having tried the demo, writes that it is “a bit cartoonish”, movement is limited to a small area with a boundary, and there are rendering errors, which the post omits. The outlet retells the physics claim more strongly than the post (“laws of physics”). The comparison with Decart’s Oasis and the figures of $230 million and a valuation above a billion are the outlet’s, not the post’s.