Xuancheng Jin · Rengan Xie · Wenting Zheng · Rui Wang · Hujun Bao · Yuchi Huo

• [Sept.15.2025] Fuse3D is accepted to SIGGRAPH Asia 2025. The code is still being organized. Stay tuned for updates!

• [Nov.30.2025] The code is publicly released!

• System: The code is tested on Linux.
• Hardware: An NVIDIA GPU with at least 16GB of memory is necessary. The code has been verified on NVIDIA A6000 GPUs.
• Software:
  • The CUDA Toolkit is needed to compile certain submodules. The code has been tested with CUDA versions 11.8.
  • Python version 3.8 or higher is required.

1. Clone the repo:

   git clone https://github.com/JINNMnm/Fuse3D.git
   cd Fuse3D

2. Install the dependencies:

   As Fuse3D builds upon TRELLIS. You can find more details about the dependencies in the TRELLIS repository.

   . ./setup.sh --new-env --basic --xformers --flash-attn --diffoctreerast --spconv --mipgaussian --kaolin --nvdiffrast

We do not modify the pretrained models of TRELLIS. The weights will be automatically downloaded when you run:

Fuse3DPipeline.from_pretrained("microsoft/TRELLIS-image-large")

Optionally, you can manually download the weights from HuggingFace and change the path in the above command to the local path.

Fuse3DPipeline.from_pretrained("path/to/local/directory")

Since our method involves manual masking, we provide an interactive Gradio demo to facilitate the process and simplify testing.

You can launch the web interface by running:

python app.py

This code builds upon TRELLIS. We sincerely thank the authors for their great work and open-sourcing the code.