A video conference endpoint includes a camera to capture video and a microphone array to sense audio. One or more preset views are defined. Images in the captured video are processed with a face detection algorithm to detect faces. Active talkers are detected from the sensed audio. The camera is controlled to capture video from the preset views, and from dynamic views created without user input and which include a dynamic overview and a dynamic close-up view. The camera is controlled to dynamically adjust each of the dynamic views to track changing positions of detected faces over time, and dynamically switch the camera between the preset views, the dynamic overview, and the dynamic close-up view over time based on positions of the detected faces and the detected active talkers relative to the preset views and the dynamic views.