GPU manufacturer NVIDIA has introduced PartPacker, an AI-based system designed to generate editable 3D models from a single 2D image. Unlike traditional methods that produce unified meshes, PartPacker creates part-based models, allowing individual components to be edited or animated separately for increased flexibility.
The system is intended to support workflows across various fields, including 3D printing, animation, gaming, and academic research, by facilitating asset creation and enabling more detailed customization.
The research behind PartPacker was developed in collaboration with Peking University and Stanford University.
Overview and Functionality of PartPacker
PartPacker generates highly detailed 3D objects from a single 2D RGB image by leveraging a dual volume packing technique within a Diffusion Transformer architecture. This approach arranges parts into interconnected volumes, each of which can be edited or animated independently for greater flexibility. The system’s network integrates a VAE alongside a rectified flow model that refines the latent codes conditioned on the input image. Unlike conventional methods, PartPacker simultaneously generates two latent codes instead of just one, improving detail and control.
PartPacker outputs 3D triangle meshes in the GLB format with resolutions up to 512³, optimized for NVIDIA GPU acceleration. This capability facilitates the creation of high-quality 3D assets suitable for games, films, and interactive media. Additionally, it supports export in popular 3D printing formats such as STL and 3MF, enabling multi-material printing workflows.
Advantages and Availability
Conventional 3D generation methods often yield monolithic meshes that restrict modification of individual parts. PartPacker addresses this limitation by producing modular, editable components, providing a flexible alternative that benefits industries requiring customizable 3D assets. This technology empowers creators and researchers to generate editable, part-based 3D models from a single image, enabling workflows that were previously challenging or unfeasible.
Researchers and developers can access PartPacker via the GitHub repository, which offers the source code and data processing scripts needed for mesh conversion. Additionally, a live demo hosted on Hugging Face enables users to upload images and instantly generate editable 3D models. Pre-trained VAE and Flow models are also available for download to support mesh reconstruction and 3D generation from images.
For more information visit NVIDIA.
AI-Powered 3D Model Tools
In 2024, Shenzhen-based 3D printer manufacturer Bambu Lab released PrintMon Maker, its new AI-powered 3D model generator. Available via MakerWorld, the platform enables users to create 3D printable characters using either text or image prompts. The generated designs can be imported directly into Bambu Lab’s slicing software and printed on its 3D printers without the need for manual adjustments. Additionally, the tool supports the creation of separate files for components like eyes and bases, offering greater flexibility and customization in the 3D printing workflow.
Last year, Nvidia created its Magic3D text-to-3D model tool in response to Google’s DreamFusion offering and Physna Inc.’s generative AI prototype for 3D models and scenes. Outlined in a research paper, Nvidia’s AI platform employs a two-stage method that creates a rough, low-resolution model which is then optimized to a higher resolution. Magic3D can also conduct prompt-based 3D mesh editing. Once a base prompt and a low-resolution 3D model are uploaded, the text can be modified to change the resulting model.
Join our Additive Manufacturing Advantage (AMAA) event on July 10th, where AM leaders from Aerospace, Space, and Defense come together to share mission-critical insights. Online and free to attend. Secure your spot now.
Who won the 2024 3D Printing Industry Awards?
Subscribe to the 3D Printing Industry newsletter to keep up with the latest 3D printing news.
You can also follow us on LinkedIn, and subscribe to the 3D Printing Industry Youtube channel to access more exclusive content.
Featured image shows Dual Volume Packing, Image via NVIDIA.
