Tencent Hunyuan3D: The First Open-Source 3D Model Supporting Both Text-to-3D and Image-to-3D 127

Full name Hunyuan3D-1.0, an open-source 3D model by Tencent that supports both text-to-3D and image-to-3D generation, addressing the limitations in speed and generalization of existing 3D generation models.

#行业圈子 #3D网站

Visit Tencent Hunyuan3D: The First Open-Source 3D Model Supporting Both Text-to-3D and Image-to-3D

Tencent Hunyuan3D: The First Open-Source 3D Model Supporting Both Text-to-3D and Image-to-3DIntroduction

Tencent Hunyuan 3D, fully named Hunyuan3D-1.0, is a 3D open-source model launched by Tencent that supports both text-to-3D and image-to-3D generation, addressing the shortcomings of existing 3D generation models in generation speed and generalization capability.

Tencent Hunyuan 3D, Hunyuan3D.webp

The Tencent Hunyuan 3D generation model is based on Diffusion technology and supports generating 3D assets from text and images. The model is equipped with carefully designed text and image encoders, a diffusion model, and a 3D decoder, enabling multi-view generation, reconstruction, and single-view generation. The Tencent Hunyuan 3D large model can rapidly generate exquisite 3D objects and is suitable for a variety of downstream applications.

Technical Features and Innovations of Tencent Hunyuan 3D:

Rapid Generation: The lightweight version of the model can generate high-quality 3D assets within 10 seconds, supporting fast single-image-to-3D generation. End-to-end generation, including mesh and texture extraction, is completed within 10 seconds.

Two-Stage Generation Method: Hunyuan3D-1.0 adopts a two-step process of multi-view generation and multi-view reconstruction. First, a multi-view diffusion model synthesizes 6 new images from different viewpoints; then, a Transformer-based sparse-view large-scale reconstruction model processes these images to ultimately generate 3D assets.

Adaptive CFG Technology: During the multi-view generation stage, the model adjusts the classifier-free guidance (CFG) scale values for different viewpoints and time steps to ensure that the generated images are both close to the input and diverse.

Comprehensive Open Source: Model weights, inference code, and algorithms are all open-sourced, with both a lightweight version and a standard version provided. The lightweight version can generate 3D assets in about 10 seconds on an A100 GPU, greatly lowering the barrier to 3D content creation.

Strong Generalization Capability: With strong generalization and controllability, it can reconstruct objects of various scales—from large buildings to small tools and plants—meeting diverse creative needs.

Tencent Hunyuan 3D Performance and Applications.gif

Performance and Applications of Tencent Hunyuan 3D:

Beyond SOTA: In qualitative and quantitative evaluations, Hunyuan3D-1.0 outperforms other open-source models, especially in terms of generation speed and generalization capability.
Wide Range of Applications: Hunyuan3D technology has been applied across multiple Tencent business scenarios. For example, the 3D car logo reconstruction speed for Tencent Maps has been improved by 91%. Tencent Maps, based on the Hunyuan 3D large model, has also launched a custom 3D navigation car logo feature, allowing users to create personalized 3D navigation car logos. Compared to traditional solutions, the speed has been improved by 91%. Additionally, the "3D Character Dream Factory" feature has been launched, supporting personalized 3D character generation and providing users with more creative freedom.
Industry Impact: The open-sourcing of Hunyuan3D-1.0 provides a powerful tool for game development, film and television production, virtual reality, and other fields, promoting innovation and application of AI-generated 3D content.

Application Scenarios of Tencent Hunyuan 3D:

Game Development: Generate high-quality 3D assets such as game characters, props, and buildings, significantly improving game development efficiency.
Film and Animation: Automatically generate 3D film characters and motion effects for creators, assisting in animation production.
E-commerce Advertising: Generate 3D product models based on advertising creative themes, achieving interactive special effects and enhancing creative ad content.
Virtual Reality (VR) / Augmented Reality (AR): Generate realistic 3D virtual environment elements and interactive content, enhancing immersive experiences.

The Tencent Hunyuan 3D generation large model "Hunyuan3D-1.0" was officially open-sourced on November 5. It is the industry's first open-source large model that supports both text and image input for 3D generation, available in both a lightweight version and a standard version. The lightweight version can generate high-quality 3D assets in just 10 seconds, suitable for the fine-tuning and deployment needs of enterprises and developers. Hunyuan3D will greatly improve the efficiency of 3D asset generation, offering developers more creative opportunities.

Github Project: https://github.com/Tencent/Hunyuan3D-1

Hugging Face: https://huggingface.co/tencent/Hunyuan3D-1