Hunyuan3D-2.1 is an industrial-grade, fully open-source 3D asset generation system developed by Tencent. Released in June 2025, it represents a significant advancement in the Hunyuan3D series by providing a comprehensive, full-chain open-source framework. The system enables the creation of high-fidelity 3D assets from single images or text prompts, releasing not only the model weights but also the training code and data processing pipelines to the public.
The model architecture consists of two primary components working in tandem: Hunyuan3D-DiT for shape generation and Hunyuan3D-Paint for texture synthesis. The shape generation stage uses a flow-based diffusion transformer combined with a high-fidelity mesh autoencoder (ShapeVAE) that leverages surface importance sampling to capture complex geometric details. The texture stage employs a mesh-conditional multi-view diffusion model that ensures view consistency and detail alignment across the generated asset.
A defining feature of version 2.1 is its Physically-Based Rendering (PBR) material synthesis. Unlike previous iterations that generated standard RGB textures, this version produces multi-channel maps—including albedo, normal, metallic, and roughness—consistent with the Disney Principled BRDF model. This capability significantly improves visual realism under varying lighting conditions, allowing for photorealistic light interaction within professional rendering engines like Unreal Engine and Blender.
Prompting and Best Practices
For the highest quality results, users are advised to provide reference images with clearly defined subjects and clean backgrounds; the model also includes a built-in background removal feature to assist with this process. A Guidance Scale in the range of 5.0 to 7.5 is recommended to balance adherence to the input image with geometric accuracy. Additionally, increasing the Step Count to 30 or more can significantly improve the resolution and clarity of the generated textures and materials.