From NVIDIA: 2024-03-21 09:00:51

NVIDIA researchers have introduced LATTE3D, a text-to-3D generative AI model capable of instantly creating 3D representations from text prompts. The model significantly speeds up the generation process, allowing for near-real-time results on a single GPU, like the NVIDIA RTX A6000. LATTE3D can be utilized for video games, design projects, and virtual training environments.

Designed to streamline the creative process, LATTE3D offers multiple 3D shape options based on text prompts, allowing creators to quickly iterate and optimize objects for higher quality. The model can export shapes to graphics software applications like NVIDIA Omniverse for Universal Scene Description-based workflows. Developers can train LATTE3D on various datasets for personalized applications.

NVIDIA Research, home to numerous scientists and engineers, showcased advancements at NVIDIA GTC. LATTE3D was trained using NVIDIA A100 GPUs and diverse text prompts generated by ChatGPT to enhance its capabilities. The model’s versatility extends to various industries, from landscape design to robotics, supporting rapid 3D shape generation and customization.

For more cutting-edge NVIDIA Research updates, visit the NVIDIA Technical Blog and explore sessions at GTC, taking place in person in San Jose, Calif., and online until March 21. Check out Jensen Huang’s GTC keynote for the latest news on NVIDIA AI developments.



Read more at NVIDIA: NVIDIA Gen AI Research Brews 3D Shapes in Under a Second