README.md

metadata

title: TRELLIS.2 Multi-Image Conditioning
emoji: 🧊
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 6.1.0
app_file: app.py
pinned: false
license: mit
short_description: Multi-view image to 3D generation

TRELLIS.2 Multi-Image Conditioning Fork

This fork extends TRELLIS.2 with multi-image conditioning and Windows support.

What's New

• Multi-image conditioning: Use multiple views for better 3D reconstruction
• Windows support: Runs on Windows with automatic sdpa attention fallback

For the interactive visualization tool, see the viser_view branch.

Installation

git clone https://github.com/OpsiClear/Trellis2_multi_image_conditioning.git --recursive
cd Trellis2_multi_image_conditioning
. ./setup.sh --new-env --basic --flash-attn --nvdiffrast --nvdiffrec --cumesh --o-voxel --flexgemm

On Windows, flash-attn is unavailable. The code automatically falls back to PyTorch's native sdpa backend.

Usage

Multi-Image Generation

from trellis2.pipelines import Trellis2ImageTo3DPipeline
from PIL import Image

pipeline = Trellis2ImageTo3DPipeline.from_pretrained("microsoft/TRELLIS.2-4B")
pipeline.cuda()

# Load multiple views
images = [Image.open(f"view_{i}.png") for i in range(4)]

# Generate with multi-image conditioning
mesh = pipeline.run_multi_image(images)[0]

Or run the example:

python example_multi_image.py

Other Examples

python example.py           # Single image generation
python app.py               # Gradio web demo
python example_texturing.py # PBR texture generation
python app_texturing.py     # Texture generation web demo

Attribution

This project is a fork of TRELLIS.2 by Microsoft Corporation, originally released under the MIT License.

For full documentation, training instructions, and model details, see the original repository.

If you use this code, please cite the original paper:

@article{xiang2025trellis2,
    title={Native and Compact Structured Latents for 3D Generation},
    author={Xiang, Jianfeng and Chen, Xiaoxue and Xu, Sicheng and Wang, Ruicheng and Lv, Zelong and Deng, Yu and Zhu, Hongyuan and Dong, Yue and Zhao, Hao and Yuan, Nicholas Jing and Yang, Jiaolong},
    journal={Tech report},
    year={2025}
}

License

The original TRELLIS.2 code is MIT licensed by Microsoft Corporation. New additions in this fork (multi-image conditioning) are licensed under AGPL-3.0.