tencent/hunyuan3d-2

Hunyuan3D 2.0 represents a significant advancement in the field of 3D content creation, providing a comprehensive large-scale 3D synthesis system designed for generating high-resolution textured 3D assets. This innovative system is built upon two core foundation models: Hunyuan3D-DiT, a sophisticated large-scale shape generation model, and Hunyuan3D-Paint, a cutting-edge large-scale texture synthesis model.

The shape generative model, Hunyuan3D-DiT, leverages a scalable flow-based diffusion transformer architecture to produce 3D geometry that accurately aligns with a given input condition image. This precise alignment forms a robust foundation for a wide range of downstream applications. Complementing this, the texture synthesis model, Hunyuan3D-Paint, capitalizes on strong geometric and diffusion priors to synthesize high-resolution and vibrant texture maps. These textures can be applied to both algorithmically generated meshes and user-created, hand-crafted meshes, offering versatility and control to the user.

To further streamline the 3D asset creation workflow, we introduce Hunyuan3D-Studio, a versatile and user-friendly production platform. Hunyuan3D-Studio simplifies the process of recreating and manipulating 3D assets, empowering both professional 3D artists and amateur creators to efficiently modify and even animate their 3D meshes.

Rigorous systematic evaluations have been conducted to assess the capabilities of Hunyuan3D 2.0. The results demonstrate that Hunyuan3D 2.0 significantly outperforms previous state-of-the-art 3D synthesis models, including both open-source and closed-source solutions, across key metrics such as geometry detail, condition alignment fidelity, and texture quality.

Hunyuan3D 2.0: Core Features and Architecture

Two-Stage Generation Pipeline

Hunyuan3D 2.0 employs an efficient two-stage generation pipeline, strategically decoupling the complexities of shape and texture synthesis. This pipeline begins with the creation of a bare 3D mesh, followed by the synthesis of a detailed texture map specifically designed for that mesh. This approach not only simplifies the generation process but also provides users with the flexibility to apply high-quality textures to both generated and pre-existing, handcrafted 3D models.

Performance Benchmarking and Results

Extensive performance evaluations have been carried out to benchmark Hunyuan3D 2.0 against leading open-source and closed-source 3D generation methodologies. The quantitative results conclusively demonstrate Hunyuan3D 2.0's superior performance in generating high-quality textured 3D assets and its enhanced ability to adhere to input conditions.

Performance Comparison Table

Visual Examples of Generation Results

The following animations showcase the generation capabilities of Hunyuan3D 2.0, demonstrating the high-fidelity and quality of the synthesized 3D assets.

Community Resources and Extensions

We extend our sincere gratitude to the community members who have contributed to the Hunyuan3D 2.0 ecosystem. Below are listed notable extensions and resources developed by the community:

• ComfyUI-3D-Pack
• ComfyUI-Hunyuan3DWrapper
• Hunyuan3D-2-for-windows
• 📦 A bundle for running on Windows | 整合包
• Hunyuan3D-2GP

Acknowledgements and Open Source Dependencies

We gratefully acknowledge the invaluable contributions of the open-source research community and the developers of the following repositories, whose work has significantly facilitated the development of Hunyuan3D 2.0:

• DINOv2
• Stable Diffusion
• FLUX
• diffusers
• HuggingFace
• CraftsMan3D
• Michelangelo