Nvidia reckons its new AI pipeline is the end of tedious prototyping for 3D artists

Prototyping can be a time-consuming process for 3D artists and developers, requiring hours of manual tweaking on models that might only end up getting scrapped. Nvidia reckons its latest AI Blueprint can help make that a thing of the past by applying generative AI to 3D modelling.

At the electronics trade show IFA Berlin today, the maker of many of the best graphics cards, launched its new AI Blueprint for 3D Object Generation. The pipeline allows artists to use text prompts to populate 3D scenes with models that can then be modified in Blender, one of the best 3D modelling software programs, if needed.

Nvidia notes that in traditional workflows, 3D artists create placeholder assets to populate prototypes of 3D scenes, adjusting core elements before visuals can be refined and finalised. A lot this is throwaway work, it says, and artists lose time on tedious modeling when they could be engaged in creative exploration.

The new customisable Nvidia AI Blueprint for 3D object generation aims to speed up the process by providing a simple pipeline through which artists can automate the prototyping process using text prompts.

The artist can start by providing an idea for a scene using a prompt. The blueprint’s built-in large language model will then generate 20 possible objects to include in the scene with the Llama 3.1 8B NVIDIA NIM microservice accelerating the results.

NVIDIA SANA, a text-to-image framework, generates previews of the potential objects, and each one can be regenerated, modified or discarded, giving the artist creative control. The assets can also be refined in Blender (automatic export is provided) or in other 3D modelling software.

Image 1 of 2

The assets can then be converted into ready-to-use 3D models via the Microsoft TRELLIS Nvidia NIM microservice, which accelerates the process. Microsoft TRELLIS is a 3D asset generation model developed by Microsoft Research that allows users to generate detailed 3D assets with complex shapes and textures using text or image prompts.

While high-quality text-to-image 3D models for consumer use can be difficult to set up, TRELLIS NIM integrates with the Nvidia AI Blueprint, simplifies deployment and accelerates generation by 20% on Nvidia RTX GPUs. Nvidia says that with PyTorch optimisations, the average time saved is approximately 6 seconds per object on an Nvidia GeForce RTX 5090 GPU (see below for today's deals).

You can learn more about the pipeline in the Nvidia AI Garage blog. If you need to upgrade your setup for 3D work, see our pick of the best laptops for 3D modelling.