In a historic milestone in the world of 3D printing, researchers at top tech company NVIDIA have introduced the innovative LATTE3D—an advanced text-to-3D AI model that’s capable of quickly (under a second!) transforming text prompts into detailed 3D representations. This means that designs can be whipped up in almost real-time on a single graphics processing unit (GPU).
LATTE3D by NVIDIA
Sanja Fidler, the VP of AI research at NVIDIA, underscored that, compared to previous similar processes that took up to an hour to generate, LATTE3D can produce high-quality 3D visuals in seconds. In order to achieve this, the company leveraged a popular rendering format and refined the AI model’s efficiency.
The program works by providing multiple shape options based on text prompts without the user having to rely on pre-existing assets or starting from scratch with each design, thereby streamlining the creative process. Users can thereafter quickly optimize selected shapes for higher quality and export them to platforms likeNVIDIA Omniverse, or graphics software.
Presentation of LATTE3D
On March 18, 2024, NVIDIA CEO Jensen Huang presented the LATTE3D at the NVIDIA GTC, showcasing its magnitude, and highlighting how these impressive and substantial AI advancements in training diffusion models are set to change and shape the future.
NVIDIA aims to empower creators and developers across several industries by continually improving AI capabilities and offering innovative tools for efficient 3D content generation.
Training LATTE3D
LATTE3D was initially trained on datasets for everyday objects and animals, however, it has an adaptable architecture which enables developers to train the model on various different data types.
Developers then trained LATTE3D using NVIDIA A100 Tensor Core GPUs and varied text prompts using ChatGPT to improve its understanding of different phrases that described specific 3D objects.
LATTE3D Applications
This innovation will be useful for a number of applications, including video game development, advertising, design products, and robotics training. This technology could aid in the development of personal assistant robots, computer graphics, computer vision, and self-driving cars. It could even come in handy for landscapers who want quick garden designs with elaborate layouts.
Image Credit: Shutterstock/photosince
