PartCrafter - NU United Bytes open source single figure 3D generated models

What is PartCrafter?

PartCrafter is an advanced 3D generative model, jointly proposed by Peking University, ByteDance and Carnegie Mellon University. It can generate multiple semantically explicit and geometrically diverse 3D mesh parts from a single RGB image at one time. The model achieves independent evolution and global consistency at the component level through combinatorial potential space and hierarchical attention mechanism. Based on a pre-trained 3D mesh diffusion transformer (DiT), PartCrafter excels in generation quality and efficiency, supporting end-to-end generation from single objects to complex scenes.

Features of PartCrafter

• Multi-part generation of single diagramsThe 3D mesh is a unique method that generates multiple semantically explicit and geometrically diverse 3D mesh parts from a single RGB image, breaking through the limitations of traditional methods.
• Combined potential space: Each 3D part is represented by a set of decoupled latent tokens, and parts can evolve independently during the generation process, preserving part-level details while ensuring overall consistency.
• Hierarchical Attention Mechanisms: Support structured information flow within individual components and between all components to ensure global consistency in the generation process and improve generation quality.
• Pre-trained model inheritance: A pre-trained 3D mesh-based diffusion transformer (DiT), which inherits the pre-trained weights, encoder and decoder, further improves the model generation capability and efficiency.
• End-to-end component-aware generation: Under the condition of a single image, PartCrafter can denoise multiple 3D parts at the same time, realizing end-to-end part-aware generation from individual objects to complex multi-object scenes.
• No need to pre-segment images: Unlike traditional methods, PartCrafter does not rely on pre-segmented images and can generate multiple parts directly from a single image, simplifying the generation process.
• High-quality generation of results: The generated 3D models have good geometric structure and visual effects, support ultra-high definition geometric detail modeling, and are suitable for a variety of application scenarios.
• Wide range of application scenarios: Wide range of applications in game development, architecture and interior design, film and television production, education, and augmented reality/virtual reality, enabling rapid generation of high-quality 3D assets.
• Open source and resourceful: The project is open source, providing the official website, GitHub repository and arXiv technical papers and other resources, support for pre-training weights to download and structure reuse, convenient for developers and researchers to use and research.

PartCrafter's core strengths

• Multi-part generation of single diagrams: Multiple 3D parts with well-defined semantics and different geometries can be generated from a single RGB image without additional input.
• No pre-segmentation required: It does not rely on pre-segmented images and generates widgets directly from raw images, simplifying the generation process and lowering the threshold of use.
• High quality geometric details: Generate 3D models with fine geometry and good visual effects, support UHD modeling.
• Component-independent evolution: By combining potential spaces, each component can evolve independently, preserving details while ensuring overall consistency.
• Hierarchical information flows: Utilize hierarchical attention mechanisms to ensure effective flow of information between and within components to improve generation quality.
• Pre-trained models help: A pre-trained 3D mesh diffusion transformer (DiT) based on inheriting advanced weights and architectures to improve generation efficiency and effectiveness.
• End-to-end generation capabilitiesIt realizes end-to-end generation from a single image to a complex 3D scene, and supports diversified application scenarios.

What is PartCrafter's official website

• Project website:: https://wgsxm.github.io/projects/partcrafter/
• Github repository:: https://github.com/wgsxm/PartCrafter
• arXiv Technical Paper:: https://arxiv.org/pdf/2506.05573

Who PartCrafter is for

• 3D artists and designers: It can quickly generate high-quality 3D models and improve the efficiency of creation, applicable to game development, film and television production, architectural visualization and other fields.
• game developer: Used to quickly generate 3D assets such as characters, props and scenes in games, accelerating the game development process and reducing production costs.
• Architects and interior designers: Helps to quickly construct architectural models and interior decorating schemes for scheme presentation and client communication, enhancing design efficiency.
• Educators and students: Used to demonstrate complex scientific concepts such as molecular structure, human anatomy, etc. to enhance teaching and learning experience.
• AR/VR Developers: Generate realistic 3D models for augmented reality and virtual reality applications to enhance user experience.
• Researchers and developers: Open source code and pre-trained models provide researchers with research and development facilities that can be used for academic research and technological innovation.