Rack focus still needs a puller a prompt cannot replace

The shot

Rack focus is the technique of shifting focus between two planes within a single continuous shot, drawing the eye from one subject to another without a cut. A cinematography guide's definition of the technique describes it as "the filmmaking technique of changing the focus of the lens during a continuous shot," moving the focal plane from one object in frame to another, and states that the shallower the depth of field, the more visible the transition becomes. The same guide notes that on a professional set the move is typically executed by a dedicated Focus Puller, working with the camera operator to adjust the focus ring live during the take.

What the documents show

The guide's definition ties rack focus to a physical lens property, depth of field, which a real camera computes from its aperture, focal length and sensor size. A generation vendor documents the same visual result differently: Runway's own camera-terms reference lists "rack focus" among a set of "focus techniques" alongside deep, soft and shallow focus, each defined by a short description and a natural-language prompt example rather than by a numeric aperture or sensor value. The documentation shows depth of field being requested as a described look, not derived from lens optics the way an actual Focus Puller's marks are.

The workflow

Specifying a rack focus for a generation brief means writing the two planes and the direction of the shift into the prompt itself, in the terms the vendor's own reference demonstrates, rather than setting an aperture or a focus distance. Because the technique is prompt-described, matching a generated rack focus to the depth-of-field character of surrounding live-action coverage is a visual comparison to make after generation, checking whether the described shift reads with a similar shallowness and pace, rather than a parameter that can be set to match in advance.

What the tool does not change

No vendor documentation reviewed here claims a computed depth-of-field control equivalent to a lens's aperture and sensor size, and this note does not extend that claim on their behalf; availability of even the prompt-level technique should be checked against each vendor's own documentation rather than assumed. The judgment a Focus Puller makes on set, timing a shift to a performance beat, is still made by a person deciding how a prompt should describe that timing, not by the tool measuring it.

• Does the prompt name both focal planes and the direction of the shift, the way the vendor's own examples do?
• Does the generated shift's pace and shallowness read as consistent with the surrounding footage's actual depth of field?
• Has this vendor's documentation been checked directly, rather than assuming a feature documented elsewhere carries over?

A Focus Puller's job was never only turning a ring on cue; it was judging when a shift in attention should happen, and that judgment still has to be written into the prompt by a person before any generation begins.