In OpenCV, when you try to do a face-recognition, the existing algorithms work with an input array, representing the image. So you hve to work with an image.
Plus, there are different ways to get a stream from a camera (like VideoCapture, or cvCaptureFromCAM depending on the language), but the only thing you can do with theses is to "take a snapshot", save this image in a matrix, and treat the image.
So yeah, there's no difference between face detection in an image and face detection from a camera