Amidst the continuous development of AI technology, Kuaishou's LivePortrait open-source project has emerged at a critical juncture. Its timing is crucial, as it coincides with significant advancements in the DiT video model. This open-source project has paved a controllable path for facial performances in AI videos, which is of great significance.
Building on this, Kijai has updated his LivePortrait ComfyUI plugin, bringing a series of remarkable achievements. Firstly, it has essentially achieved the functionality of real-time facial expression transfer via camera. This means users can capture their expressions through the camera and then transfer these expressions to other objects or scenes in real-time, adding more fun and creativity to video creation. Secondly, it has also achieved the functionality of transferring expressions to existing videos. This is a very practical tool for video post-production, allowing creators to change the emotional expressions of people in the video without re-shooting, thus giving the video new life.
Moreover, during the update process, Kijai has also optimized and adjusted some key parts of the original project. Among them, the original project's face detection model, Insightface, has been replaced with the commercially viable Google MediaPipe. Google MediaPipe has higher accuracy and stability, and can better adapt to different application scenarios and environmental conditions. This replacement not only enhances the performance of the plugin but also makes it more viable and promising for commercial applications.