The rapid growth of the live streaming industry brought about the VTuber trend, where content creators use character avatars to stream. One of the most accessible way to move a character in real time is by using a vision-based facial motion capture technique. However, previous works still suffer from jittering issues, which hinder the quality of the character’s movements. This work aims to develop a smoothed facial motion capture system that works with a live2D VTuber model. The system combines Mediapipe Face Mesh and OpenCV solutions to capture facial landmarks, which are then used to calculate head pose estimation using the Perspective-n-Point (PnP) function. In addition, the system uses EAR and MAR functions to detect facial features. The motion values obtained from this process are then filtered using a Kalman filter. Finally, the filtered motion data is sent to the Unity engine, which drives the Live2D VTuber model by adjusting the character’s motion parameters. The developed system successfully captures and drives the Live2D VTuber model with smoother motion, overcoming the jitter problem prevalent in previous facial motion capture approaches. The system’s improved motion capture quality makes it a more viable option for a wide range of potential uses.