Nvidia has announced a new artificial intelligence model called LATTE3D for producing three-dimensional shapes.
The new model converts text into realistic three-dimensional sculptures of objects and animals in a very short time.
This model provides generated shapes in a common format that can be used in standard graphics modeling applications, allowing them to be incorporated into virtual environments.
The results of the model align with several fields such as video games, advertising campaigns, design projects, virtual robotics training grounds, and virtual reality worlds, as reported by Nvidia researchers.
LATTE3D represents a significant advancement, as advanced models for converting text into three-dimensional images just a year ago required an hour to produce images of the same quality.
While models previously could do this in only 12 seconds, creators across various industries can now convert text into three-dimensional shapes in real-time.
According to Nvidia, the new model can produce detailed three-dimensional shapes almost instantly when run on a single graphics processing unit, such as the Nvidia RTX A6000.
LATTE3D is changing the game for designers, who can now use the model to instantly generate multiple shape ideas instead of spending a long time searching for three-dimensional assets in the library.
A small set of three-dimensional shapes is created for each text message, giving the designer multiple options to choose from.
The chosen design can then be refined and the graphics quality improved within a few minutes using textual commands.
When satisfied with the results, the designer can export them to a graphic programming application such as Nvidia’s Omniverse. This platform supports the USD framework for exchanging three-dimensional graphic data.
Nvidia trained LATTE3D on a database containing shapes of animals and everyday objects, and researchers can adjust the training to include different types of data such as buildings or vehicles to expand its capabilities.
The model was trained using Nvidia’s older A100 Tensor Core GPU units, and ChatGPT from OpenAI was used to teach the model to understand a variety of textual commands, enhancing its ability to comprehend different phrases used to describe three-dimensional objects.
