Skip to content

JoyHallo

A digital human model for Mandarin video generation

Visit official site

What JoyHallo is

The model is described as a digital human model designed for Mandarin video generation, built from the jdh-Hallo dataset of 29 hours of Mandarin video collected from JD Health International employees. The dataset covers different ages and speaking styles including dialogue and professional medical topics, and the model uses a Chinese wav2vec2 model for audio feature embedding, with inference stated as 14.3% faster.

What you can do with it

  • Generate video content from audio
  • Work on complex lip movements in Mandarin
  • Generate video in both English and Mandarin
  • Capture relations between lip, expression and pose features with a semi-decoupled structure

Application scenarios

  • Generate educational videos for language learning
  • Produce professional medical education videos
  • Create entertainment videos to add variety to content work

Getting started

  • Visit the official JoyHallo website
  • Read the product introduction and feature notes
  • Download and install the required software or plugin
  • Import or record an audio file to prepare video generation

Pros & cons

✓ What we like

  • The jdh-Hallo dataset covers different ages and speaking styles
  • A semi-decoupled structure relates lip, expression and pose features

! What to watch out for

  • The model is designed around Mandarin video generation
  • Audio feature embedding depends on the Chinese wav2vec2 model

FAQ

What data is the model built on?

The material names the jdh-Hallo dataset of 29 hours of Mandarin video from JD Health International employees.

Can it generate English video?

The feature list states both English and Mandarin video generation are supported.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review