Michelangelo is ready, Nvidia introduces Neurangelo: an AI model that extracts 3D objects from 2D videos.

Whether Nvidia’s new AI model will also master the many talents of a specific Renaissance artist remains to be seen, but with the Neurangelo designation, the famous chipmaker raises some expectations. The new system generates 3D objects from 2D videos.

A modern master

Neurangelo creates 3D replicas of objects from 2D videos. For example vehicles, statues or buildings, just to name a few options. This is even possible with videos created with a smartphone.

The model selects several frames from the video in which the respective object is visible from different angles. This gives the system a clear picture of perspective, volume and shape. First, a raw 3D image is sketched and then the details are tweaked.

Neurangelo builds on Nvidia’s previous Instant NeRF 3D technology to work out the necessary details of a print. According to the developers at Nvidia, the newcomer performs significantly better with textures that are difficult to display, such as marble or roof tiles. The company also claims that Neurangelo can recreate a 3D scene in around two hours, which is a blast.

Nvidia is also indirectly involved in another innovative 3D project: OpenAI’s Shap-E runs on an NVIDIA V100 GPU. Incidentally, the need for GPUs for AI has not harmed the American chip company, both sales and the average price have recently risen sharply.