1. Tutorial Introduction

HunyuanWorld-Mirror is an open-source 3D world generation model released by Tencent's Hunyuan team in October 2025. It supports multiple input methods, including multi-view images and videos, and can output various 3D geometric prediction results such as point clouds, depth maps, and camera parameters. The model adopts a pure feedforward architecture, can be deployed on a single GPU, and achieves second-level inference by processing 8-32 view inputs locally. Its technical architecture includes multimodal prior hints, a general geometric prediction architecture, and a curriculum learning strategy. Through a dynamic prior injection mechanism, the model can flexibly adapt to any combination of priors. During training, a curriculum learning strategy based on task order, data scheduling, and progressive resolution is employed to maximize generalization ability. HunyuanWorld-Mirror performs excellently in 3D point cloud reconstruction and end-to-end 3DGS reconstruction, demonstrating outstanding geometric accuracy and detail restoration capabilities. Related research papers are available at [link to relevant paper].WorldMirror: Universal 3D World Reconstruction with Any-Prior Prompting".

This tutorial uses a single RTX 5090 graphics card as the default resource, but a minimum single RTX 4090 graphics card can be used to start the program.

2. Effect display

3. Operation steps

1. Start the container

2. Usage steps

If "Bad Gateway" is displayed, it means the model is initializing. Since the model is large, please wait about 2-3 minutes and refresh the page.  The first reconstruction takes about 2 minutes, and loading data and displaying the 3D effects may take some time.

Parameter options:

• Video Sample Interval: The interval between video sampling sessions.
• Reconstruction Options (not applied to 3DGS):
  • Show Camera: Whether to display the camera.
  • Show Mesh: Whether to display the Mesh.
  • Filter low confidence & depth/normal edges: Whether to filter low confidence and depth/normal edges.
  • Filter Sky Background: Whether to filter the sky background.
• 3D Gaussian Splatting: An interactive 3D Gaussian sputtering visualization with RGB and depth video (available for download as PLY).
• Point Cloud/Mesh: Point cloud.
• Depth Maps: Single-view depth estimation results (downloadable in PNG format).
• Normal Maps: Per-view surface orientation visualization (downloadable as PNG).
• Camera Parameters: Estimated camera pose and intrinsic parameters (downloadable as JSON).

4. Discussion

🖌️ If you see a high-quality project, please leave a message in the background to recommend it! In addition, we have also established a tutorial exchange group. Welcome friends to scan the QR code and remark [SD Tutorial] to join the group to discuss various technical issues and share application effects↓

Citation Information

The citation information for this project is as follows:

@article{liu2025worldmirror,
  title={WorldMirror: Universal 3D World Reconstruction with Any-Prior Prompting},
  author={Liu, Yifan and Min, Zhiyuan and Wang, Zhenwei and Wu, Junta and Wang, Tengfei and Yuan, Yixuan and Luo, Yawei and Guo, Chunchao},
  journal={arXiv preprint arXiv:2510.10726},
  year={2025}
}