fudan-generative-vision/hallo3
A Video Diffusion Transformer that animates static portrait images into dynamic, realistic videos driven by reference inputs.
Not currently ranked — collecting fresh signals.
star history
Hallo3 is a portrait image animation system that uses a Video Diffusion Transformer architecture to generate highly dynamic and realistic video sequences from a single portrait image. The model takes a driving input (such as another video or audio) to control the motion and expressions of the portrait subject. It represents a state-of-the-art approach to talking head generation and character animation published at CVPR 2025.
Frequently asked
- What is fudan-generative-vision/hallo3?
- A Video Diffusion Transformer that animates static portrait images into dynamic, realistic videos driven by reference inputs.
- Is hallo3 open source?
- Yes — fudan-generative-vision/hallo3 is open source, released under the MIT license.
- What language is hallo3 written in?
- fudan-generative-vision/hallo3 is primarily written in Python.
- How popular is hallo3?
- fudan-generative-vision/hallo3 has 1.4k stars on GitHub.
- Where can I find hallo3?
- fudan-generative-vision/hallo3 is on GitHub at https://github.com/fudan-generative-vision/hallo3.