NVIDIA Research has unveiled an AI model that could become a tool for the metaverse.
This AI, which has been trained on a million 2D images taken from different angles, can generate 3D objects and characters. So entire cities could be built in the metaverse implementing this dynamic.
This is the new NVIDIA AI that can create 3D objects from images
NVIDIA’s new AI model could become an essential tool for creating virtual worlds, without the need for specific knowledge or a lot of time:
GET3D can generate a virtually unlimited number of 3D shapes based on the data it is trained on. Like an artist turning a lump of clay into a detailed sculpture, the model transforms numbers into complex 3D shapes.
For example, just training the AI ​​with 2D images of chairs can generate different models and known varieties. The larger the database for AI training, the more variety and detail we’ll see in the results, says the NVIDIA team.
A dynamic that could be useful for creating the virtual worlds of video games or building in the metaverse. And one of the peculiarities offered by this NVIDIA proposal is that the objects created can be edited to adapt to the function of the project.
Incorporating another NVIDIA Research AI tool, StyleGAN-NADA, developers can use text prompts to add a specific style to an image, such as modifying a rendered car to become a burnt-out car or taxi, or converting a normal house in an enchanted one.
And this is just the beginning, as NVIDIA plans a new version of this AI that could, for example, be trained on real-world data.
