Hunyuan3D 2.0 is a comprehensive system for generating high-resolution, textured 3D assets from input images. It targets both professional and amateur 3D creators, offering a two-stage pipeline that excels in geometry detail, condition alignment, and texture quality, outperforming existing state-of-the-art models.

How It Works

The system employs a two-stage generation pipeline: first, a shape generation model (Hunyuan3D-DiT) creates a base mesh using a scalable flow-based diffusion transformer. Second, a texture synthesis model (Hunyuan3D-Paint) generates high-resolution texture maps for the mesh, leveraging diffusion priors. This decoupling allows for flexibility in texturing both generated and custom meshes.

Quick Start & Requirements

• Installation: pip install -r requirements.txt, followed by pip install -e . and custom compilation for rasterizer/renderer.
• Prerequisites: PyTorch (install via official site), Python.
• Hardware: 6 GB VRAM for shape generation, 16 GB VRAM for shape and texture generation.
• Resources: Official site, Hugging Face Spaces, Gradio Apps, API Server, Blender Addon.

Highlighted Details

• Outperforms top open-source and closed-source models in CMMD, FID_CLIP, FID, and CLIP-score benchmarks.
• Offers various model sizes (0.6B, 1.1B) and versions (Turbo, Fast, Multiview).
• Integrates with Blender via an addon and provides a Gradio app and API server for easy deployment.
• Supports Windows, macOS, and Linux.

Maintenance & Community

• Active development with frequent releases (e.g., Turbo models, texture enhancement).
• Community support via Wechat and Discord.
• Links to community extensions like ComfyUI wrappers and Windows installers.

Licensing & Compatibility

• Apache 2.0 License.
• Permissive for commercial use and integration with closed-source projects.

Limitations & Caveats

The project is actively developed with recent releases, indicating potential for breaking changes. A TensorRT version is planned but not yet released.