Model Card for PartPacker
Description
PartPacker is a three-dimensional (3D) generation model that is able to generate part-level 3D objects from single-view images. We introduce a dual volume packing strategy that organizes all parts into two complementary volumes, allowing for the creation of complete and interleaved parts that assemble into the final object. This model is ready for non-commercial use.
License/Terms of Use
Deployment Geography
Global
Use Case
PartPacker takes a single input image and generates a 3D shape with an arbitrary number of complete parts. Each part can be separated and edited independently to facilitate downstream tasks such as editing and animation.
It's intended to be used by researchers and academics to develop new 3D generation methods.
Release Date
- Github: 06/11/2025 via https://github.com/NVlabs/PartPacker
- Huggingface: 06/11/2025 via https://huggingface.co/NVlabs/PartPacker
Reference(s)
Model Architecture
Architecture Type: Transformer
Network Architecture: Diffusion Transformer (DiT)
Input
Input Type(s): Image
Input Format(s): Red, Green, Blue (RGB)
Input Parameters: Two-dimensional (2D) image
Other Properties Related to Input: Resolution will be resized to 518x518.
Output
Output Type(s): Triangle Mesh
Output Format: GL Transmission Format Binary (GLB)
Output Parameters: Three-dimensional (3D) triangle mesh
Other Properties Related to Output: Extracted at a resolution up to 512^3; without texture.
Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA’s hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions.
Software Integration
Runtime Engine(s)
- PyTorch
Supported Hardware Microarchitecture Compatibility
- NVIDIA Ampere
- NVIDIA Hopper
Preferred Operating System(s)
- Linux
Model Version(s)
v1.0
Training, Testing, and Evaluation Datasets
We perform training, testing, and evaluation on the Objaverse-XL dataset. For the VAE model, we use the first 253K meshes for training and the rest 1K meshes for validation. For the Flow model, we use all 254K meshes for training.
Objaverse-XL
Link: https://objaverse.allenai.org/
Data Collection Method: Hybrid: Automatic, Synthetic
Labeling Method by dataset: N/A (no labels)
Properties: We use about 254k mesh data, which is a subset from the Objaverse-XL filtered by the number of parts.
Inference
Acceleration Engine: PyTorch
Test Hardware: NVIDIA A100 (1 GPU configuration)
Ethical Considerations
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
Please report security vulnerabilities or NVIDIA AI Concerns here.