We will consider the following two broad categories :
Real world content
By this is meant the use of still photography or video/film techniques to capture natural scenes (including studio techniques such as filming elements against green/blue screen). One technique is to use a single camera that is moved horizontally between shots- this can be used when the scene to be captured is static. A more generally usable scenario is to use a pair of identical cameras, mounted in such a way to ensure that their optical axes are horizontally separated by an amount that is appropriate for the distance from the cameras to the target object.
In this context, each camera will acquire the visual content to be presented to the viewer's corresponding eye during later stereoscopic display and viewing.
The camera pair will record the left and right images simultaneously. The level of synchnronism may not be perfect, but it should be sufficient that there is no time-based difference between the two left/right images large enough to spoil the stereo effect when the images are viewed together. How the cameras are mounted depends on a number of factors, including the camera body size and shape, how close the cameras' optical axes need to be, is there a need for rapid variation of interaxial separation, etc. Mounting methods range from simple mounting bars using fixed holes, to motorised mirror-based rigs that allow arbitrary axial separation with a large variety of camera bodies.
Imaginary / non-real content
By this is meant the development, by computer-generated imagery or other techniques, of images that are not directly derived from imaging real world scenes. This could be the result of employing ray-tracing software such as POV-Ray, 3DS, Maya and a host of others; or even by carefully planned and executed drawing or painting. In a CGI example, one might render an image pair for a virtual scene, with different virtual camera locations in each scene; the virtual cameras for the virtual scene are then analogues of a real-world camera pair.
In this context, separate renders of a scene are used to generate the visual content to be presented to the viewer's corresponding eye during later stereoscopic display and viewing.
No matter how the content is obtained, the image pairs must somehow be presented in an appropriate way to the viewer. This will be discussed briefly in the next FAQ entry.