Instant Skinned Gaussian Avatars
for Web, Mobile and VR Applications

SUI 2025 Demo Track

Naruya Kondo1, Yuto Asano1, Yoichi Ochiai1,
1University of Tsukuba, Japan
Paper arXiv Code Demo

We propose an instant avatar generation system and a portable avatar format.
Our avatar generation and viewer run entirely in the browser!

💃 Create and enjoy your 3D avatar — Super easy! 💃
(a) Scan yourself using Scaniverse (≈3 min)
(b) Export PLY data (Share > Export > PLY)
(c) Visit our demo and select your data (≈30 sec)
(d) Done!
All processing runs on your device. Try it with your friends!

Visit our demo site for a real-time view!

Avatar Generation

Visit our demo site to try it out!

Portable Avatar Format: Gaussian-VRM (.gvrm)

For more details, check out the GitHub repository.

How Gaussian-VRM Works

In the background, a mesh (in the .vrm format) runs within GVRM,
and the poses of the splats follow the vertices of the mesh. That's all! :)

Technical note for expert readers:

We implemented the core logic in GLSL to achieve high performance. However, as you know, depth sorting of splats is the most computationally expensive part. Updating the positions of all splats every frame for the sorter process would cause excessive memory I/O, making real-time rendering nearly impossible. So, we divide the splat scene into 14 sub-scenes and use the static positions within each sub-scene for sorting. In other words, we do not use the skinned results for sorting, but we do use the correct skinned results for rendering. This approach allows us to achieve a realistic computational cost while maintaining visual accuracy, even in WebGL. ✌

Abstract

We present Instant Skinned Gaussian Avatars, a real-time and cross-platform 3D avatar system. Many approaches have been proposed to animate Gaussian Splatting, but they often require camera arrays, long preprocessing times, or high-end GPUs. Some methods attempt to convert Gaussian Splatting into mesh-based representations, achieving lightweight performance but sacrificing visual fidelity. In contrast, our system efficiently animates Gaussian Splatting by leveraging parallel splat-wise processing to dynamically follow the underlying skinned mesh in real time while preserving high visual fidelity. From smartphone-based 3D scanning to on-device preprocessing, the entire process takes just around five minutes, with the avatar generation step itself completed in only about 30 seconds. Our system enables users to instantly transform their real-world appearance into a 3D avatar, making it ideal for seamless integration with social media and metaverse applications.

Video

Applications

Acknowledgements

This work was supported by the Ochiai Pavilion at the Osaka/Kansai Expo 2025.
This work was supported by JSPS KAKENHI Grant Number 23KJ0284.

BibTeX

@misc{kondo2025instantskinnedgaussianavatars,
      title={Instant Skinned Gaussian Avatars for Web, Mobile and VR Applications}, 
      author={Naruya Kondo and Yuto Asano and Yoichi Ochiai},
      year={2025},
      eprint={2510.13978},
      archivePrefix={arXiv},
      primaryClass={cs.CG},
      url={https://arxiv.org/abs/2510.13978},
}