AB
AiBoss
project

ReSyncer - An AI video editing tool jointly launched by Tsinghua University and Baidu

ReSyncer is an AI video editing tool jointly developed by Tsinghua University and Baidu. It generates high-quality lip-motion videos synchronized with sound through audio-driven processing. ReSyncer uses Style-SyncFormer to analyze sound and create 3D facial models...

What is ReSyncer?

ReSyncer is an AI video editing tool jointly developed by Tsinghua University and Baidu. It generates high-quality lip-sync videos that are synchronized with the audio, driven by sound. ReSyncer uses Style-SyncFormer to analyze the sound and create a 3D facial model, combining it with the target video to generate a synchronized and expressive virtual character. ReSyncer supports personalized fine-tuning, speech style switching, and face-swapping functions, making it suitable for scenarios such as virtual host and performer creation, and live streaming. It excels in synchronizing audiovisual facial information.

ReSyncer's main functions

  • Lip-syncGenerates lip movements synchronized with the given audio.
  • Style transfer: Transfer specific speaking styles or facial expressions to target videos.
  • Personalized fine-tuning: Quickly adjust the generated facial animation to match the facial features of a specific person.
  • Video-driven lip-syncUse facial images from the target video to drive lip-sync animation.
  • face-swapping technologyReplace one person's facial features with another person's for identity transformation or special effects.

ReSyncer's technical principles

  • 3D facial model generationUsing Style-SyncFormer, a deep learning model, to predict 3D facial dynamics based on vocal features.
  • Stylized facial animationIt learns stylized 3D facial dynamics through the Transformer architecture, achieving precise synchronization of facial expressions and lip movements.
  • Style-based generatorsThe predicted 3D facial dynamics are combined with facial images in the target video to generate high-fidelity facial images.
  • Facial feature fusionDuring the generation process, a simple insertion mechanism is used to fuse 3D facial mesh information with stylized features, thereby improving the quality and stability of lip synchronization.

ReSyncer project address

Application scenarios of ReSyncer

  • Film and video productionIn film and video production, ReSyncer can be used to achieve complex special effects, such as face swapping or lip-syncing, to increase visual appeal.
  • Advertising industryIn advertising production, style transfer functionality can be used to create unique visual effects and attract viewers' attention.
  • Social media and content creationContent creators can use ReSyncer to enhance their video content, such as creating funny parody videos using face-swapping technology.
  • Education and trainingIn language learning or professional training, lip-syncing can help learners better understand and imitate pronunciation.