What Is a Rack Focus?

What it does to attention

Every shot tells the audience where to look, usually through composition, movement or light. A rack focus does it with the lens, and it does it as a transfer rather than a statement. Something is sharp, then something else is, and the eye follows because sharpness is the strongest attention cue in a frame.

That transfer carries meaning that a cut cannot. A cut says "now look here." A focus pull says "these two things are in the same room, and the relationship between them just changed." Which is why it is the standard tool for a character noticing something behind them, an object becoming important, or an eavesdropper being revealed in the background of a conversation.

The direction matters. Pulling from foreground to background tends to feel like discovery or dread, because you are being shown something you had not registered. Pulling from background to foreground feels like a conclusion landing, or attention snapping back to the person in front of you.

The optics it needs

Three conditions have to be true, and if any one of them is missing the pull simply will not be visible.

• Shallow depth of field. Longer focal length, wider aperture, or both. Deep focus makes the technique impossible by definition.
• Depth separation. The two subjects must sit at genuinely different distances. Two people side by side cannot be racked between.
• Something to read as sharp. Texture, edges, printed type, eyes. A flat painted wall does not register a focus change even when it happens.

On set the fourth condition is a focus puller with marked distances, because at f/1.4 the margin for error is a few centimetres. This is why the technique is a rehearsal-dependent shot rather than a spontaneous one.

Variants worth knowing

A snap focus covers the same distance in a handful of frames and reads as impact. A rack across three planes pulls to a middle subject and continues, which is showy and hard to keep clean. A split diopter avoids the pull entirely by keeping both planes sharp at once, and is the right answer when you want simultaneity instead of transfer.

Prompting a rack focus in AI video

This is one of the harder camera effects to get out of a video model, because the model has no lens and no focus distance. It is producing what a focus change looks like, statistically, from the description you give it. That means the description has to supply the geometry the optics would normally supply.

• Name both subjects and both distances explicitly: a ringing telephone sharp in the foreground, a woman blurred in the doorway behind it. If you only name one subject, there is nothing for the plane to travel to.
• State the direction and the timing in one clause: focus shifts from the telephone to the woman halfway through the shot. Vague phrasings like changing focus or dreamy focus usually give you an overall blur ramp.
• Include lens language, because it correlates strongly with the look in training data: 85mm at f/1.8, very shallow depth of field, bokeh.
• Lock the camera. Adding static camera prevents the model from substituting a dolly or a zoom, which is the most common misread of the request.

Failure modes, roughly in order of frequency: the whole frame softens and then re-sharpens without either subject ever being isolated; the model performs a slow zoom instead of a focus change; the background subject changes identity or clothing mid-pull because it was never sharp enough to be committed to; and the pull happens instantly on the first frame, so the shot has no before state.

Two workarounds actually hold up. Generate the shot twice with identical prompts and different focus states, then cross-dissolve the two clips over 12 to 20 frames, which produces a clean pull because both end states are stable. Or generate everything deep-focus and add the blur ramp in post with a mask, which gives you exact control over speed and is usually faster than re-rolling a rack focus until the model cooperates.