Shooting with stereoscopic depth as compositional element—convergence, interaxial distance, planes of focus calculated per shot. Gravity-like restraint, not novelty.
Stereoscopic vision on screen only works if you place two cameras side-by-side at an exact distance—this distance is called the baseline or eye separation, and it's your first decision. It determines how aggressively the sense of depth appears. For a Hollywood drama, I use a 65 mm baseline; for extreme close-up action, I reduce it to 45 mm, because otherwise the viewer's eyes become strained and headaches occur. This isn't theory—you'll see it in the faces of the audience after the third viewing.
The convergence point—the spot where both camera views intersect—is your compositional tool. If you place it on the actress, she is in the picture plane; if you place it behind her, the focus shoots into the negative (pop-out effect); if you place it closer to the camera, depth is built into the image. This is deliberate direction with geometry. On set, you calibrate both cameras to pixel accuracy—a deviation of 2 pixels creates visible ghosting, i.e., double images, in the cinema. That's why you need not only a DIT but also a stereoscopic supervisor who is constantly measuring.
3D is not a gimmick if you plan it from the screenplay onwards. Avatar showed us how: Cameron didn't think in terms of two cameras, but in layers of depth as a narrative device. The Pandora landscape is resolved into layers—foreground, midground, background—and thus tells a more spatial story. If, on the other hand, you convert a 2D production to 3D retrospectively, it appears artificial because the lighting, focus, and compositing decisions were not intended for depth.
In practice: For flat scenes (dialogue, close-ups), I shoot with a large convergence distance—the depth is subtle. Action scenes or landscapes require an aggressive baseline and active convergence ramps (this is the movement of the convergence point during the shot). Edits must be made carefully—the convergence jump between two shots must not exceed 2–3 diopters, otherwise the eye tires. This is technically as strict as classic editing, but with an additional geometric dimension. Your monitor on set shows you both images superimposed—you have to learn to read this overlay information like axis lines. It's a new language for the camera, and it demands respect for your viewers' eyes.