In traditional 3D workflows, artists start with placeholder, low-fidelity assets to block out a scene. These rough models are later refined, detailed, and finalized. But the process is slow, often involving tedious, throwaway work that leaves less time for creative exploration.
Generative AI can help by suggesting draft objects and automating intermediate steps. However, connecting multiple AI models into a usable pipeline has historically been complex.
NVIDIA AI Blueprints solve this by offering pre-built workflows that simplify deployment and customization. Today, NVIDIA introduced the AI Blueprint for 3D Object Generation, which allows artists to generate up to 20 objects for a scene from a simple text prompt.
The blueprint also integrates the new Microsoft TRELLIS NVIDIA NIM microservice, which produces high-quality 3D assets 20% faster than the native model.
Generate, Prototype, Repeat
Every 3D project begins with an idea — followed by decisions about theme, style, location, textures, and more. Even after assets are placed, scenes often go through multiple rounds of iteration.
The new AI Blueprint streamlines this cycle. Artists provide a text prompt describing their concept, and a built-in large language model (LLM) generates up to 20 object ideas. Powered by the Llama 3.1 8B NVIDIA NIM microservice, this stage delivers results quickly while suggesting creative prompts for less-experienced users.
Preview images of these objects are generated by NVIDIA SANA, a text-to-image framework that produces high-resolution renderings. Artists can refine, regenerate, or discard previews before converting selected ones into 3D models using the Microsoft TRELLIS NIM microservice.
TRELLIS not only accelerates model generation by 20% but also creates highly detailed objects with complex shapes and textures, ready for use in game design, architecture, and digital media. Completed assets are automatically exported to Blender for refinement, with options to move them into other popular 3D platforms.
Streamlined Deployment
Setting up an equivalent workflow manually would require time, technical expertise, and experimentation. By packaging the required models and optimizations, NVIDIA AI Blueprints simplify deployment on GeForce RTX and RTX PRO GPUs.
With PyTorch optimizations, the TRELLIS NIM microservice saves an average of six seconds per object on an NVIDIA GeForce RTX 5090 GPU — a huge time saver for freelancers or studios generating hundreds of assets across projects. The service is supported on all RTX 50 Series and 40 Series GPUs (desktop and laptop) with 16GB or more memory.
Ready, Aim, Deploy
Artists can get started with the AI Blueprint for 3D Object Generation in just a few steps:
- Load the blueprint with the included models.
- Enter a scene prompt (e.g., “night market” or “sunny beach day”).
- Review 20 automatically generated preview images.
- Replace, edit, or delete objects as needed.
- Convert previews to 3D models individually or all at once.
- Export assets to Blender for further refinement.
Detailed instructions and resources are available on NVIDIA Build and GitHub.
Explore More
NVIDIA continues to expand its collection of AI Blueprints, including workflows for 3D-guided generative AI and other NIM microservices.
For those attending the IFA trade show in Berlin, join the session “How AI Is Changing Content Creation” on Saturday, Sept. 6, at 2 p.m. CET, featuring Sean Kilbride, NVIDIA’s director of professional visualization technical marketing, alongside other industry experts.
