NVIDIA populates virtual worlds

The Future. Software company NVIDIA is releasing an AI-powered object-generator that can help fill virtual worlds with an endless array of vehicles, animals, and other miscellanea to make them feel lived in. With the ability to change objects with simple text commands, the software could become a go-to resource for unlocking the imagination of even the most amateur developers.

AI into existence

Engadget reports that NVIDIA is in the business of bringing virtual worlds to life with a few keystrokes.

• NVIDIA is releasing an AI software called GET3D, which can “generate characters, buildings, vehicles and other types of 3D objects.”

• And the software does it very fast, pumping out 20 objects per second on a single computer.

The software has been trained on vast datasets similar to those used by AI-generative art tools like Midjourney, DALL-E, and Stable Diffusion — except only using “synthetic” images. The hope is that the models can one day be trained on “real-world” images.

And by using it with another of NVIDIA’s AI tools, StyleGAN-NADA, the users could manipulate the creations via text prompts, like turning a mansion into a haunted house.

No dead spaces

The ultimate ambition of GET3D is for developers to be able to populate vast virtual worlds easier than ever before. That could include…

• Metaverse spaces: a virtual world will only be interesting if it feels as spontaneously filled as our real one.

• Video games: as games become borderless, developers will need a way to fill the gaps with things to interact with quickly.

NVIDIA also notes that the software could be helpful in robotics and architecture — two industries that require a hefty amount of previsualization.