Auto reframe

The technique combines face or subject detection with a smoothed camera path. Software finds the subject in each frame, decides where a vertical window should sit to contain it, then smooths that path over time so the crop glides rather than jittering with every small movement. Shot changes are detected separately so the window can jump cleanly at a cut instead of panning across it.

It matters because almost every long form source is horizontal and every short form feed is vertical. A fixed centre crop loses two thirds of the width and routinely pushes a speaker who sits off centre half out of frame. Manual keyframing works but costs minutes per clip, which does not scale to a daily posting schedule. Auto reframe makes the conversion a one click step.

The nuance is in the hard cases. Two people in one shot force a decision between a split layout, switching between them, or a wide crop that keeps both small, and the right answer depends on who is talking. Fast motion, screen shares, text on screen, and subjects with no face can confuse detection, and over eager tracking produces a frame that swims. Good tools expose the choice of layout and let the path be corrected.

In practice the workflow is reframe first, then caption, because text has to be placed against the final vertical frame. Checking the result on a phone sized preview catches the moments where the crop lands on the wrong person or clips a gesture, and a speaker aware reframe that follows whoever is talking removes most of those corrections.