How AI Picks the Vertical Layout for Each Shot

A vertical window cut from a 16:9 frame is only about a third of its width. That's enough for one person talking to the camera, but not for two people sitting apart, a person pointing at a whiteboard, or a full-screen slide.

Most long videos change shots every few seconds, so a clip inherits several of them. Choosing per shot means each one gets the layout its content needs, and shots that show the same thing get the same answer, so the clip doesn't jump between layouts for no reason.

The goal for every shot is the same: the phone viewer should follow the clip as easily as possible. To judge that, the AI gets more than pixels:

• The clip's title, description and what is said in it, so it knows whether the clip is about the speaker, the guest's reaction or the chart on screen.
• One frame from each shot and how long that shot stays on screen.
• What is known about the whole video, such as a webcam box that stays in the same corner.
• Faces found in each shot, so a window lands on the person rather than next to them.

A split screen halves the size of everything in it, and every switch between split and single is visible to the viewer. So the AI uses one window by default and splits only when two people are actually talking in the same wide shot.

A cut to the other person isn't a reason to split: the next shot simply gets its own window. Someone walking through the background isn't either. A content panel next to a person is used only when the video has a webcam or avatar presenter, or an overlay edited into the picture. Side-by-side comparisons and edited-in graphics are shown as a whole frame so nothing important is cropped away.

Within a shot, the window holds still. A frame that drifts after every head movement is harder to watch than a steady one, so each shot gets one window that covers where the subject moves.

When the subject is wider than a vertical window, such as a person with what they're holding, or someone bent over a tablet, the window doesn't narrow to a column or jump to a stray face. It widens to hold the whole subject, and the picture is shown slightly smaller over a blurred copy of itself. You get thin blurred bars instead of a cropped product or a missing hand.

The AI's choice is a starting point. If you'd frame a shot differently, change it in the Clip Editor:

• Faces are fully in frame, including on short shots of a second or two.
• Text on slides and screens is readable on a phone-sized preview.
• Captions don't cover the face or the bottom window of a split.
• The clip doesn't switch layouts more than the content does.
• Objects people show or point at are in the picture, not just beside it.