Hunyuan3D 2.0

Hunyuan3D 2.0 is a large-scale 3D synthesis system designed to generate high-resolution textured 3D assets. It consists of two primary foundation components:

1. Hunyuan3D-DiT: A large-scale shape generation model built on a scalable flow-based diffusion transformer. It creates geometry that aligns with a provided condition image.
2. Hunyuan3D-Paint: A large-scale texture synthesis model that produces high-resolution, vibrant texture maps for both generated and hand-crafted meshes.

The system is designed to decouple the complexities of shape and texture generation, providing flexibility for various 3D asset creation workflows.

Hunyuan3D 2.0 is a large-scale 3D synthesis system designed for generating high-resolution textured 3D assets. It is composed of four primary foundation components:

1. Hunyuan3D-DiT: A large-scale model dedicated to shape generation.
2. Hunyuan3D-Paint: A large-scale model dedicated to texture synthesis.
3. Hunyuan3D-Studio: A production platform for manipulating or animating meshes, suitable for both professional and amateur workflows.
4. FlashVDM: A universal acceleration framework used to speed up processes.

Hunyuan3D 2.0 utilizes a two-stage generation pipeline to create 3D assets:

1. Stage 1: Shape Generation: The system first creates a 'bare mesh' (geometry without textures) using the shape generation model.
2. Stage 2: Texture Synthesis: The system then synthesizes a texture map for that specific mesh.

This decoupled approach allows users to either use the full end-to-end pipeline or apply the texture synthesis model to existing, hand-crafted meshes.

New users can get oriented with ∇-Prox through the following resources:

• Quick Tour: A guided overview to become familiar with the features and functionalities of ∇-Prox.
• API Reference: A complete technical list of all available classes and functions.
• Tutorials: In-depth guides covering advanced topics and best practices.

When running Hunyuan3D 2.0, ensure your hardware meets the following VRAM requirements:

• Shape generation only: Requires at least 6 GB VRAM.
• Shape and texture generation: Requires at least 24.5 GB VRAM in total.

You can host a local Gradio interface for Hunyuan3D 2.0 using the gradio_app.py script. Depending on the model variant you wish to use (mini, multi-view, or standard), you must specify the correct --model_path and --subfolder. All commands below include --low_vram_mode to optimize memory usage.

# Hunyuan3D-2mini
python3 gradio_app.py --model_path tencent/Hunyuan3D-2mini --subfolder hunyuan3d-dit-v2-mini --texgen_model_path tencent/Hunyuan3D-2 --low_vram_mode

# Hunyuan3D-2mv
python3 gradio_app.py --model_path tencent/Hunyuan3D-2mv --subfolder hunyuan3d-dit-v2-mv --texgen_model_path tencent/Hunyuan3D-2 --low_vram_mode

# Hunyuan3D-2
python3 gradio_app.py --model_path tencent/Hunyuan3D-2 --subfolder hunyuan3d-dit-v2-0 --texgen_model_path tencent/Hunyuan3D-2 --low_vram_mode

Use an editable installation if you need to use the main version of the source code or if you intend to test changes made to the code.

Important: You must retain the cloned DeltaProx folder to continue using the library after an editable installation.

git clone git+https://github.com/princeton-computational-imaging/Delta-Prox.git
cd DeltaProx
pip install -e .

You can install dprox directly from PyPI using pip. Ensure you have PyTorch installed beforehand by following the official PyTorch installation instructions.

pip install dprox

You can use Hunyuan3D 2.0 directly within Blender by using the provided Blender Addon (blender_addon.py).

Prerequisite: You must have an API server launched and running before using the addon in Blender.

Installation and Usage:

1. Obtain the blender_addon.py file from the repository.
2. Follow the specific tutorial provided in the repository to install and configure the addon within your Blender environment.