VoxHammer: Training-Free Precise and Coherent 3D Editing in Native 3D Space

🏠 Project Page | Paper | Edit3D-Bench | Online Demo (Comming soon)

TL;DR: A training-free 3D editing approach that performs precise and coherent editing in native 3D latent space instead of multi-view space.

🔥 Updates

• [2025-08-27] Release inference code and Edit3D-Bench.

📦 Installation

Please set up the Python environment following TRELLIS, or you can create by:

Prerequisites

• System: Linux (Ubuntu 20.04/22.04 recommended)
• GPU: NVIDIA GPU with at least 40 GB memory (an A100 80GB is strongly recommended)
• Python: 3.10
• Conda (Miniconda/Anaconda) recommended

Installation Steps

1. Create and activate the environment

conda create -n hammer python=3.10 -y
conda activate hammer

2. Install requirements from requirements.txt

pip install -r requirements.txt

3. Install requirements from GitHub projects

pip install git+https://github.com/huanngzh/bpy-renderer.git
pip install git+https://github.com/NVlabs/nvdiffrast.git
pip install git+https://github.com/EasternJournalist/utils3d.git@9a4eb15e4021b67b12c460c7057d642626897ec8
git clone https://github.com/autonomousvision/mip-splatting.git /tmp/extensions/mip-splatting
pip install /tmp/extensions/mip-splatting/submodules/diff-gaussian-rasterization/

💡 Usage

Step 1 — Add encoder to pipeline

After downloading "TRELLIS-image-large" model, we can find "sparse_structure_encoder" in ckpts/ss_enc_conv3d_16l8_fp16.safetensors and "slat_encoder" in ckpts/slat_enc_swin8_B_64l8_fp16.safetensors.

Before using these two encoders, we should add {"sparse_structure_encoder": "ckpts/ss_enc_conv3d_16l8_fp16", "slat_encoder": "ckpts/slat_enc_swin8_B_64l8_fp16"} in pipeline.json, then the pipeline can correctly call them.

Step 2 — Create a 3D mask

Prepare a mesh that marks the editable region (e.g. assets/example/mask.glb).

mask_prepare.mp4

Inputs required (example):

• assets/example/model.glb — original 3D asset
• assets/example/mask.glb — mask mesh

Step 3 — Render RGB views & masks

python utils/render_rgb_and_mask.py \
--source_model assets/example/model.glb \
--mask_model assets/example/mask.glb \
--output_dir outputs

Step 4 — Inpaint the rendered views

python utils/inpaint.py \
--image_path outputs/images/render_0002.png \
--mask_path outputs/images/mask_0002.png \
--output_dir outputs/images \
--prompt "A dog."

Step 5 — Run inference

python inference.py \
--input_model assets/example/model.glb \
--mask_model assets/example/mask.glb \
--image_dir outputs/images \
--output_dir outputs

Evaluation

Explore Edit3D-Bench for more details.

License

This project is licensed under the MIT License.
However, please note that the code in trellis originates from the TRELLIS project and remains subject to its original license terms.
Users must comply with the licensing requirements of TRELLIS when using or redistributing that portion of the code.

Citation

@article{li2025voxhammer,
    title = {VoxHammer: Training-Free Precise and Coherent 3D Editing in Native 3D Space},
    author = {Li, Lin and Huang, Zehuan and Feng, Haoran and Zhuang, Gengxiong and Chen, Rui and Guo, Chunchao and Sheng, Lu},
    journal = {arXiv preprint arXiv:2508.19247},
    year = {2025}
}