Neuralangelo Research Reconstructs 3D Scenes | NVIDIA vs Text-To-4D

Neuralangelo is an NVIDIA Research model that reconstructs detailed 3D objects and scenes from ordinary 2D video. Feed clips filmed from multiple angles and the system builds textured meshes of buildings, sculptures, vehicles, and interiors suitable for games, digital twins, and robotics sims.

Earlier neural reconstruction methods blurred fine textures like roof shingles, glass, and marble. Neuralangelo extends instant neural graphics primitives from NVIDIA Instant NeRF to capture those surfaces with higher fidelity, so smartphone footage can produce assets creative teams can import into design tools without a laser scan or manual modeling pass.

Researchers, game artists, and robotics engineers use Neuralangelo to digitize real-world objects from Michelangelo-scale statues to campus buildings. The project is open source on GitHub under nvlabs/neuralangelo with no commercial license fees for the research code itself.

Text-To-4D is a Meta AI research project (MAV3D) that generates three-dimensional dynamic scenes from text descriptions. The method uses a 4D dynamic Neural Radiance Field optimized for appearance, density, and motion consistency by querying a text-to-video diffusion model.

Unlike static 3D generators that output a single mesh, Text-To-4D produces scenes you can view from any camera angle and composite into other 3D environments. The approach needs no 3D or 4D training data; the underlying text-to-video model trains only on text-image pairs and unlabeled videos.

The project page hosts demo samples for text-to-4D prompts like "a corgi playing with a ball" and image-to-4D conversions from still photos. It is a research showcase, not a commercial product with a public API or signup. Researchers and 3D artists interested in NeRF-based dynamic scene generation can explore the paper and sample outputs on the site.