Horizontal to Vertical Video: Smart Face-Tracking Crop, Not a Flat Center Crop

You have a horizontal clip: an interview, a webinar, a chunk of a podcast, something shot on a camera. The platforms want vertical. Reels, Shorts and TikTok play full-screen on a phone, and that shape is 9:16. The fastest way to convert it is a center crop. The fastest way to ruin it is also a center crop.

Let's break down why a flat center crop cuts off exactly the thing that matters, how smart face-tracking crop actually works, and what numbers sit behind the claim that vertical genuinely wins attention on mobile.

Why a center crop is a bad idea

When you take a 16:9 frame and cut a 9:16 strip out of it, you physically throw away the sides. Only about 56% of the original width survives; the rest goes off-frame (Transcriptr). If you park the crop window in the center and never move it, anyone standing off-center loses half their face. The speaker steps toward the whiteboard and vanishes. Two people in a conversation become an empty wall between them.

A center crop knows nothing about the content. It cuts out the geometric middle of the rectangle, and the important part of a video almost never sits dead center.

What happens when 16:9 becomes 9:16

It helps to see the scale of the loss. A horizontal 1920x1080 frame and a vertical 1080x1920 frame are different worlds. To get true 9:16 with no black bars, you keep a narrow vertical column out of the wide frame. By width, that's a little over half of the original picture.

That's why the choice of where to place the column decides everything. Place it in the center at random and you throw away a face instead of empty background. Place it by content and glide it with the speaker, and you throw away the background while the person stays in frame.

How smart face-tracking crop works

Under the hood, a proper auto-reframe isn't magic, it's a clear pipeline. First it finds faces on every frame (for example with MediaPipe Face Mesh, which places up to 468 points per face), then decides which face is the main subject right now. Next it computes the vertical window so the face sits in frame with 15-20% headroom on top, and finally it smooths the path of that window so the shot doesn't jitter (Transcriptr).

The key step here is smoothing. If the window chases every tiny head movement, the viewer gets seasick in three seconds. So the path runs through a filter (low-pass or Kalman) with a smoothing window of roughly 10-30 frames: big moves get tracked, small shakes get damped out (Transcriptr).

• Detection: where the faces are on each frame
• Tracking: the same person across frames, so the window doesn't jump
• Crop: a narrow vertical window with air above the head
• Smoothing: a gentle camera move with no snaps

The numbers: why vertical actually wins

This isn't a matter of taste. People hold phones upright and watch that way. By figures widely cited across the industry, about 94% of smartphone users hold their phone vertically while watching video, and most won't rotate it for a horizontal clip (Embryo). A vertical frame also takes up roughly 78% more screen area on mobile than a horizontal one (Sculpt).

The same logic drives completion. A widely cited MediaBrix study found vertical videos see roughly a 90% higher completion rate than horizontal ones (Sculpt). And by recent platform observations, a clip in true 9:16 on vertical feeds pulls about 2.3x more engagement than a 1:1 square or a 4:5 cut (quso.ai). The format alone already moves the numbers.

And one thing people keep forgetting: about 85% of feed video is watched with the sound off (Sculpt). So the vertical format alone only does half the job. Without captions burned into the frame, the viewer never learns what the clip is about and scrolls on.

Center crop, manual keyframes, and tracking

There are three ways to get vertical out of horizontal. What separates them is how much time you spend and how reliably the face stays in frame.

Manual keyframing gives a perfect result if you're willing to sit and drag the crop window by hand across the whole clip. For a minute of talking that's dozens of adjustments. Smart tracking gets nearly the same quality in seconds, because an algorithm drives the window.

Framing hits retention directly

The first seconds decide. On vertical feeds the viewer scrolls endlessly, and if the opening frame has a forehead sliced off or the person already drifted off the edge, the thumb is already moving. Good framing keeps the eye exactly where it should be for the whole clip.

9:16 safe zones

Even a perfectly framed face can be ruined by the platform's interface. TikTok, Reels and Shorts draw their own buttons, handle, description and captions on top of your video. The standard canvas is 1080x1920, 9:16. Keep the important stuff away from the edges the interface covers (quso.ai).

• Bottom: keep face and text above the bottom ~270 px, where the handle, caption and music sit
• Right: leave ~100 px on the right for the like, comment and share buttons
• Top: don't push captions to the very top, where icons and the clock live
• Clips where key content spills outside the safe zone lose completion, by one estimate around 22% (quso.ai)

How Monty does it

Monty takes one take of yours and builds a finished short for each platform out of it. It writes the script in your voice, edits the take (cuts, music, b-roll), adds captions timed to your speech, makes a separate version for each platform, and posts it to Shorts, Reels, TikTok and Telegram. The first video is free.

For horizontal-to-vertical, that means you don't hand-drag a crop window or fit yourself into safe zones. Monty produces the vertical version with captions burned into the frame and posts it itself. Want control? Leave approval on and every video waits for your go. Want autopilot? Put it on a schedule.