PERSONA (ICCV 2025)

Project Page | Paper | Video

• PERSONA is a personalized whole-body animatable avatar creation system that supports pose-driven deformations (i.e., non-rigid deformations of clothes) from a single image.

Compared to previous 3D-based methods, our PERSONA better represents non-rigid deformations of clothes. For more high-resolution demo videos, please visit our website.

Compared to previous diffusion-based methods, our PERSONA better preserves identity of the person. For more high-resolution demo videos, please visit our website.

Install

• To install a conda environment and necessary packages, run below.

conda env create -f environment.yml
conda activate persona
pip install -r requirements.txt

• Also, to download and install third-party modules, first, please get granted to access SVD and sapiens pose.
• Then, login HuggingFace on your machine by running huggingface-cli login.
• Finally, run below.

bash install.sh

Creating and animating avatars from a single image

1. To create an avatar, you first need to fit SMPL-X to an image and generate training videos. Go to ./preprocess.
2. Then, go to ./avatar to create and animate the avatar.

Tracking 3D whole-body poses from a video

• To animate your avatar, you need to track 3D whole-body poses from a video. Go to ./pose_track.

Reference

@inproceedings{sim2025persona,
  title={{PERSONA}: Personalized Whole-Body {3D} Avatar with Pose-Driven Deformations from a Single Image},
  author = {Sim, Geonhee and Moon, Gyeongsik},  
  booktitle={ICCV},
  year={2025}
}