AI Reframe: turning horizontal footage vertical without beheading anyone

The single most common reason a good horizontal clip performs badly as a Short: the crop. A static center-crop from 16:9 to 9:16 throws away two-thirds of the frame — usually the two-thirds containing the speaker's face the moment they shift in their chair.

What AI reframing actually does

Modern reframing runs face detection across the timeline, builds a motion path for the active speaker, then smooths that path so the virtual camera glides instead of jittering. Good implementations add dominant-speaker logic for multi-person shots — the crop follows whoever is talking, not whoever is biggest.

When to trust it

• Talking-head and podcast footage: near-perfect, ship it.
• Two-person interviews: good, if the tool cuts between speakers rather than compromising in the middle.
• Screen recordings and gameplay: don't reframe — use a split-screen layout instead, footage on top, camera or gameplay below.

The overrides worth making

Trust the tracking, but audit three moments: the first frame (your thumbnail crop), any moment a prop or screen matters, and the final frame. A ten-second manual nudge on those beats re-rendering the whole clip. Keep eyelines in the upper third — vertical viewers' attention starts there and captions live in the lower third.

Getting started

Drop any 16:9 video into Clipdify, pick a vertical layout, and the reframe runs automatically with face tracking — then fine-tune with the crop controls if a moment needs it. The whole point: vertical stops being a re-edit and becomes an export setting.

Under the hood: how the tracking decides

A good reframe pass runs in three stages. Detection finds every face per frame with a confidence score. Tracking links detections across frames into per-person paths, surviving brief occlusions like a raised coffee cup. Smoothing then fits a damped camera path through the target's positions — the damping is the craft, because raw tracking data jitters at the pixel level, and an undamped crop gives viewers motion sickness. Clipdify's implementation adds dominant-speaker weighting on top: when audio says the left person is talking, the camera commits to them instead of hovering anxiously between faces.

Fixing the three classic failure cases

• The walk-through: someone crosses behind the speaker and the crop briefly chases them. Fix: pin the subject for that segment; ten seconds of manual override.
• The prop moment: the speaker holds up a product and the face-locked crop cuts it off. Fix: keyframe a wider crop for the reveal, then return to tracking.
• The whiteboard problem: the content is on a surface beside the speaker. Fix: alternate deliberate static crops — face, board, face — rather than asking tracking to split the difference.

Reframe vs. layout: choosing the right vertical strategy

Reframing answers 'where should the camera look?' — the right question for footage of people. Layouts answer 'how do I stack two sources?' — the right question for gameplay plus facecam, screen shares plus reaction, or podcast video plus quote cards. If you're fighting the reframer on non-face content, you're using the wrong tool: switch to a split-screen or picture-in-picture layout and give each source its own honest region.

A 5-minute quality checklist before export

1. 1Scrub at 2x speed watching only the framing — drifts jump out at speed.
2. 2Check the first frame works as a thumbnail: subject visible, no mid-blink.
3. 3Confirm captions clear the subject's chin at their lowest point.
4. 4Watch the final two seconds — endings sag when the camera path relaxes early.