RASA: Replace Anyone, Say Anything – A Training-Free Framework for Audio-Driven and Universal Portrait Video Editing

Tianrui Pan^1,2*, Lin Liu^2✉, Jie Liu^1✉, Xiaopeng Zhang², Jie Tang¹, Gangshan Wu¹, Qi Tian^2
1State Key Laboratory for Novel Software Technology, Nanjing University;
²Huawei Inc.

[Paper]      [Code]     

LIP EDITING RESULTS

RASA half-body results_1

RASA half-body results_2

MuseTalk silent parse

RASA silent parse

APPEARANCE EDITING RESULTS

source video_1

RASA appearance_editing_1

source video_2

RASA appearance_editing_2

EXTENDED APPEARANCE EDITING

Liveportrait head rotation

RASA head rotation

Liveportrait expression adjustment

RASA expression adjustment

OTHER APPEARANCE EDITING

RASA appearance_editing_1

RASA appearance_editing_2

RASA appearance_editing + singing

Our project page is borrowed from DreamBooth and ID-Animator.