3D artists often spend significant time on low-fidelity, throwaway modeling during prototyping. NVIDIA released the Artificial Intelligence Blueprint for 3D object generation to streamline that stage, providing a prebuilt workflow that generates up to 20 3D objects from a simple text prompt. The blueprint packages models and deployment components so users can skip complex setup and tailor the pipeline for their needs, and it targets both NVIDIA GeForce RTX and RTX PRO GPUs.
The pipeline starts with a built-in large language model that brainstorms object ideas from a scene prompt. The blueprint uses the Llama 3.1 8B NVIDIA NIM microservice to accelerate idea generation and NVIDIA SANA to synthesize high-resolution image previews for each object. Artists can regenerate, edit or discard individual previews. Each preview can then be converted into a ready-to-use 3D model, and the collection of up to 20 assets is primed for instant export. The blueprint includes a built-in export to Blender and supports other common 3D applications for further refinement.
Conversion from previews to 3D models is handled by the Microsoft TRELLIS NVIDIA NIM microservice, which NVIDIA says is 20% faster than the native model due to PyTorch optimizations. NVIDIA provides an example performance figure of roughly 6 seconds saved per object on an NVIDIA GeForce RTX 5090 GPU. The Microsoft TRELLIS NIM microservice is supported on GeForce RTX 50 Series and 40 Series desktop and laptop GPUs with 16GB or more. Detailed setup steps, example files and source code are available on NVIDIA Build and GitHub, and the blueprint includes in-app controls for regenerating, editing prompts, converting individual previews to 3D and exporting completed assets to Blender.
