PartCrafter
github.com
- Free plan
- Windows
- Linux
- Self-hosted
What PartCrafter is
PartCrafter is an open-source 3D generative model that creates multiple 3D parts and objects from a single RGB image. It can handle individual objects or complex scenes and produces distinct meshes with meaningful part structure without needing pre-segmented images. The project provides local inference scripts installed from its GitHub repository in a Python environment, along with pretrained object-level and scene-level weights hosted on Hugging Face. The scripts support background removal and optional style transfer to adapt real-world photos toward the rendered-image domain used for training. An optional vision-language feature can suggest the number of parts; Gemini is the default provider, and the provider architecture can be extended. The repository lists an MIT license and says the software is provided as-is, without warranty. Local inference requires a CUDA-enabled GPU with at least 8 GB of VRAM. The maintainers say they tested installation on Debian 12 with NVIDIA H20 GPUs; Windows setup guidance comes from the community. The linked Hugging Face demo is paused, and its page says users can ask the authors to restart it. The training scripts are described as suited to eight NVIDIA H20 GPUs with 96 GB of VRAM each.
Who it suits
PartCrafter is for developers and 3D practitioners who want to generate part-level objects or scenes from images and can run local inference. Its stated GPU requirement and Python-based installation are relevant to prospective users.
Good to know, and what to watch
- Generates multiple distinct 3D parts from one image.
- Supports complex scenes and individual objects.
- Pretrained object and scene weights are available.
- MIT-licensed and free.
- Optional background removal and style transfer.
- Inference needs a CUDA GPU with at least 8 GB VRAM.
- Linked Hugging Face demo is paused.
- Training scripts are suited to eight NVIDIA H20 GPUs.
- Software is provided without warranty.
Plans
Every plan and pricePartCrafter (MIT-licensed open-source software)
Free plan
The verdict
PartCrafter provides image-to-3D generation with part-aware outputs and local inference scripts. The GPU requirement, paused demo, and as-is license terms are important constraints to consider before setting it up.
What the maker publishes
- Architecture
- Its method uses disentangled latent tokens for parts and a hierarchical attention mechanism for information flow within and across parts. source
- Compositional architecture
- The model represents each 3D part with disentangled latent tokens and uses hierarchical attention across parts. source
- Demo availability
- The linked Hugging Face demo is paused, and its page says users can ask its authors to restart it. source
- Deployment
- The official instructions install the project from its GitHub repository in a Python environment and run inference scripts locally. source
- Hardware requirement
- The README specifies a CUDA-enabled GPU with at least 8 GB of VRAM for inference. source
- Image preprocessing
- The inference scripts support background removal and optional style transfer to adapt real-world photos to the rendered-image domain used for training. source
- Image styling
- An optional Gemini-powered style-transfer step converts real-world photos toward an Objaverse-style rendering before generation. source
- Integrations
- The README documents Gemini for part suggestions and style transfer, and says the provider architecture can be extended with other providers. source
- License
- The repository lists an MIT license. source
- License and warranty
- The repository uses the MIT License and provides the software “AS IS,” without warranty. source
- Model weights
- Pretrained object-level and scene-level model weights are hosted on Hugging Face. source
- Object generation
- The repository provides an inference script for generating part-level 3D objects from images. source
- Part-aware output
- It generates semantically meaningful, geometrically distinct 3D meshes without requiring pre-segmented input images. source
- Part suggestions
- An optional VLM feature can suggest the number of parts, using Gemini by default. source
- Purpose
- PartCrafter is an open-source 3D generative model that creates multiple 3D parts and objects from a single RGB image. source
- Purpose
- PartCrafter generates multiple structured 3D parts and objects from a single RGB image in one pass. source
- Scene generation
- The model supports image-to-3D generation for complex scenes as well as individual objects. source
- Scene generation
- The repository provides a separate inference script for generating 3D scenes from images. source
- Security and compliance
- The official pages opened provide no security certification or compliance claim. source
- Support
- The repository invites users to contact the maintainer by email or open an issue with questions or suggestions. source
- Tested environment
- The maintainers say they tested installation on Debian 12 with NVIDIA H20 GPUs and provide community-contributed Windows setup guidance. source
- Tested setup
- The project says its installation was tested on Debian 12 with NVIDIA H20 GPUs and points Windows users to a community guide and fork. source
- Training resources
- The project describes training scripts suited to 8 NVIDIA H20 GPUs with 96GB VRAM each. source
- Training scale
- The training scripts are described as suitable for eight NVIDIA H20 GPUs with 96 GB of VRAM each. source
Checked against the maker's pages on 9 Oct 2026.