Zhen-Dong/Magic-Me
A diffusion-based video generation system that creates custom videos featuring specific identities from provided photos using trained ID embeddings.
Not currently ranked — collecting fresh signals.
star history
The project provides a framework called Magic-Me for generating personalized videos using a pre-trained ID token. It employs a customized video diffusion (VCD) approach with three key components: an ID module trained with cropped identity using prompt-to-segmentation to separate identity from background noise; a text-to-video module with 3D Gaussian noise prior for inter-frame consistency; and video-to-video modules for face deblurring and upscaling to higher resolution.
Frequently asked
- What is Zhen-Dong/Magic-Me?
- A diffusion-based video generation system that creates custom videos featuring specific identities from provided photos using trained ID embeddings.
- Is Magic-Me open source?
- Yes — Zhen-Dong/Magic-Me is open source, released under the Apache-2.0 license.
- What language is Magic-Me written in?
- Zhen-Dong/Magic-Me is primarily written in Python.
- How popular is Magic-Me?
- Zhen-Dong/Magic-Me has 459 stars on GitHub.
- Where can I find Magic-Me?
- Zhen-Dong/Magic-Me is on GitHub at https://github.com/Zhen-Dong/Magic-Me.