MagicMirror: ID-Preserved Video Generation in Video Diffusion Transformers (ICCV 2025)

<p align="center"> <img style="margin: 20 auto; height: 140px; display: block;" src="assets/logo.png" > </p> <p align="center"> <a href='https://julianjuaner.github.io/projects/MagicMirror/'><img src='https://img.shields.io/badge/project_page-aa55dd'></a> <a href='https://arxiv.org/abs/2501.03931'><img src='https://img.shields.io/badge/arXiv_paper-ee7744'></a> </p>

ShowCases

Please refer to the project page for full-quality and more examples.

1. Reference-Driven Identity-Aware Text-to-Video Generation

| Reference Image | Generated Video | Generated Video | | ---- | ---- | ---- | | <img height="180" src="assets/teaser_6.png" > | <img height="180" src="assets/teaser_6_1.gif"> | <img height="180" src="assets/teaser_6_2.gif"> | | <img height="180" src="assets/teaser_5.png"> | <img height="180" src="assets/teaser_5_1.gif"> | <img height="180" src="assets/teaser_5_2.gif"> | | <img height="180" src="assets/teaser_4.png"> | <img height="180" src="assets/teaser_4_1.gif"> | <img height="180" src="assets/teaser_4_2.gif"> | | <img height="180" src="assets/teaser_1.png"> | <img height="180" src="assets/teaser_1_1.gif"> | <img height="180" src="assets/teaser_1_2.gif"> |

2. Stylized & Special Effect

| Reference Image | Generated Video | Reference Image | Generated Video | | ---- | ---- | ---- |---- | |<img height="128" src="assets/1.png" >| <img height="128" src="assets/1_1.gif"> | <img height="128" src="assets/2.png" >| <img height="128" src="assets/2_1.gif"> | |<img height="128" src="assets/3.png" >| <img height="128" src="assets/3_1.gif"> | <img height="128" src="assets/4.png" >| <img height="128" src="assets/4_1.gif"> | |<img height="128" src="assets/5.jpg" >| <img height="128" src="assets/5_1.gif"> | <img height="128" src="assets/5+.png" >| <img height="128" src="assets/5+.gif"> |

3. Multi-Shot Generation

A bearded man, wearing a yellow T-shirt, working for a wooden table... | Reference:<br><img height="160" src="assets/8.png" >| <img height="160" src="assets/8_1.gif"> | <img height="160" src="assets/8_2.gif"> | | ---- | ---- | ---- | | <img height="160" src="assets/8_3.gif"> | <img height="160" src="assets/8_4.gif"> | <img height="160" src="assets/8_5.gif"> |

<br>

A woman, wearing a white shirt and blue jeans, enjoying her daytime activities... | Reference:<br> <img height="160" src="assets/teaser_5.png" >| <img height="160" src="assets/6_1.gif"> | <img height="160" src="assets/6_2.gif"> | | ---- | ---- | ---- | | <img height="160" src="assets/6_3.gif"> | <img height="160" src="assets/6_4.gif"> | <img height="160" src="assets/6_5.gif"> |

Overview

• Magic Mirror: ID-Preserved Video Generation in Video Diffusion Transformers
• ShowCases
  • 1. Reference Driven Identity-Aware Text to Video Generation
  • 2. Stylized & Special Effect
  • 3. Multi-Shot Generation
• Overview
• MileStones
• Methods
• Cite Magic Mirror

MileStones

• [X] 20250101 Paper released!
• [ ] 202501-202502 We will release code and model (we are working on fit our methods on CogVideoX-1.5, HunyuanVideo, .etc). Stay tuned!

Methods

In this work, we presented Magic Mirror, a zero-shot framework for identity-preserved video generation. Magic Mirror incorporates dual facial embeddings and Conditional Adaptive Normalization (CAN) into DiT-based architectures. Our approach enables robust identity preservation and stable training convergence. Extensive experiments demonstrate that Magic Mirror generates high-quality personalized videos while maintaining identity consistency from a single reference image, outperforming existing methods across multiple benchmarks and human evaluations.

Cite Magic Mirror

If you find this repo useful for your research, please consider citing the paper

@article{zhang2025magic,
  title={Magic Mirror: ID-Preserved Video Generation in Video Diffusion Transformers},
  author={Zhang, Yuechen and Liu, Yaoyang and Xia, Bin and Peng, Bohao and Yan, Zexin and Lo, Eric and Jia, Jiaya},
  journal={arXiv preprint arXiv:2501.03931},
  year={2025}
}