Skip to content

Glossary

What is face tracking?

By the KeyMo team · Updated

Face tracking is software that finds a face in each video frame and follows it as it moves. When a wide 16:9 shot is cut to vertical 9:16, the crop follows the face so the speaker stays in frame. It works well with one person on camera and struggles with fast cuts and several speakers.

What it does when you cut to vertical

A 16:9 frame is wide, and a 9:16 frame is tall. If the source is 1920 by 1080, a vertical crop at full height is about 608 pixels wide, which is roughly 32% of the width. That is a narrow slice. If the speaker leans, walks or shifts, a fixed crop loses them. Face tracking moves the crop window to follow the face instead.

KeyMo exports 1080 by 1920 at 30 fps, and its reframe is automatic and face-tracking. The practical result is that you do not set crop keyframes by hand for a talking-head clip.

Face tracking and nearby terms

TermWhat it means
Face detectionFinding a face in a single frame
Face trackingFollowing a face from frame to frame
Face recognitionWorking out whose face it is. Tracking does not need this
Auto reframeThe whole job of choosing a crop for a new aspect ratio, which often uses face tracking
Camera or gimbal face trackingHardware that physically turns to follow a face, such as on a tripod or gimbal; it changes the shot as it is filmed, not afterwards

Where face tracking struggles

  • Several people in frame. The tool must pick one, and it may pick the wrong one or move between them.
  • Fast cuts. A multicam edit changes angle, so the tracking has to reset at each cut.
  • A face that turns away, is covered by a hand or a microphone, or is small and far from the camera.
  • Fast movement, where the crop can lag or feel jittery.
  • Cases where the person speaking is not the one the crop follows. See diarization for the audio side.

A decision rule

  1. One person, one camera, facing forward: trust the automatic crop and watch each clip once.
  2. Two people at a table in a wide shot: check that the crop is on whoever is speaking, and expect to fix some clips.
  3. Multicam edit: check each cut, since the framing changes with every angle.
  4. Moving subject or wide room: consider a different moment, or shoot closer next time with lead room in mind.

Where KeyMo fits

KeyMo returns up to 15 scored, vertical clips from one video, each with word-by-word captions in one of 12 styles and silence trimmed. If a crop is off, open the clip in the chat editor and describe what to change. Read more about aspect ratios and how a two-shot affects framing.

FAQ

What is face tracking on a camera or gimbal?

It is hardware that detects a face and moves a motor to keep it centred while you film. It acts during shooting, unlike software face tracking, which moves a crop afterwards in editing.

What is the difference between face tracking and auto reframe?

Auto reframe is the whole task of fitting a video to a new shape, such as 9:16. Face tracking is one method it can use to decide where the crop should sit.

Does face tracking work with two people talking?

Sometimes. The tool has to choose a face, and in a wide shot of two people it may stay on one or switch. Check the result, especially at turn changes.

Is face tracking the same as face recognition?

No. Tracking follows a face from frame to frame. Recognition decides whose face it is. A crop that follows a speaker does not need to know who they are.

Why does the crop drift away from the speaker?

Common causes are a face turned away, a hand or microphone over it, several people in frame, or a cut to another camera. Picking a steadier moment often fixes it.