NVIDIA has demoed a new generative AI technology called LATTE3D that can transform text prompts into 3D shapes in less than a second.
Like many other generative tools, LATTE3D quickly translates text descriptions into 3D objects that are compatible with standard rendering applications and can be integrated into virtual environments. According to Sanja Fidler, NVIDIA’s Vice President of AI Research, this represents a major improvement in efficiency, making near-real-time text-to-3D generation a practical tool for creators in various fields.
The model facilitates a rapid design process by allowing creators to generate multiple 3D shape options from a single text prompt, which can then be refined and exported for use in graphics software or platforms like NVIDIA Omniverse.
LATTE3D has been trained on datasets of animals and everyday objects, but its architecture allows for adaptation to other types of data. The model’s training included diverse text prompts generated with ChatGPT to enhance its ability to interpret a wide range of descriptive language accurately.
For more information, visit NVIDIA’s website.
