Hi, the rendering of any image (be it from a video or from a static image) on top of an image target is always done through a texture, using OpenGL;
the code is not hidden inside the Vuforia library, it is fully exposed in our samples;
for instance, have a look at the renderFrame function in the ImageTargets.cpp of the Image Targets sample, and you will find some useful geenral purpose code for rendering augmentations in OpenGL;
then if you look at the VideoPlayback sample you will find once again some OpenGL code very similar to the one used in Image Targets sample, but more focussed on the rendering of a textured rectangle (instead of a Teapot 3D model); you will also find that the texture are filled and updated with content coming from a video; the code handling the video playback is also completely exposed in the sample and is not encapsulated within the QCAR library.
So, it is really a matter of digging into our samples and understanding the code (same for the Cloud Reco sample);
one important thing is that everything related to rendering is done via OpenGL ES (i.e. Vuforia does not offer a specific 3D engine, but simply relies on OpenGL and explains how to best use it through the sample code); so, obviously you are required to master OpenGL a bi, in ordert to handle everything related to the rendering part.
I hope this helps.