Ben Mildenhall Shows Text-to-Image-to-3D Pipeline

Ben Mildenhall, a computer vision researcher and co-creator of NeRF who now works on 3D AI at World Labs, posted a reply containing a video. The clip presents a workflow that starts with a text prompt, produces an image, and then builds a 3D scene described as consistent through an atlas-based approach. A split-screen format shows rendered output alongside camera orbits around the resulting model. The post is presented as Mildenhall's own demonstration of the method rather than an external claim. No further details on training data, release plans, or performance metrics appear in the available lines.