Editing 9:16 Shorts for Retention: the Hook Frame, Caption Rhythm and the Loop Point

A vertical short is lost or kept in the first 1.5 seconds and at the last 1.5. Everything between them is caption rhythm and cut cadence — three things you can set before the first render.

· 6 min read · Omong Studio

In this guide

Most vertical videos are not rejected. They are abandoned — mid-sentence, without a comment, at a point the platform records and the editor never looks at. The graph is flat and falls, and the usual fix is "make it more interesting", which is not an instruction.

Retention in 9:16 is narrower than that. A viewer decides twice: once in the first second and a half, where nothing has been promised yet, and once at the very end, where the video either closes or loops. Everything between those two decisions is rhythm — captions, cuts and where the frame changes. All three are set before the first render, not rescued in the edit.

What each second has to do

A vertical edit is short enough to plan in beats rather than in shots. The table below is the skeleton used for a 20–30 second short; it is a budget, not a formula.

TimeJob of that windowWhat is on screen
0.0–0.5 sStop the scrollA frame that is already mid-action; no logo, no fade, no title
0.5–1.5 sShow the changeOne motion beat — a hand enters, the object moves, the shot cuts
1.5–3.0 sState the promiseThe claim in six words or fewer, spoken and captioned at once
3–8 sDeliver the first proofThe result, before the method; a number or a visible before/after
8–20 sEscalateDetail that only makes sense once the result is known
20–26 sClose the ideaRestate the promise as a fact, not a teaser
last 1.5 sSet up the loopThe closing frame returns to the composition of the opening frame

Two of those rows matter more than the rest, and neither is the middle.

The hook frame is a decision, not a title card

The first rendered frame decides whether the rest exists for that viewer. Three things reliably fail there: a black frame while audio starts, a logo, and a wide shot that has not established what is being looked at. All three ask the viewer to wait, and a swipe costs nothing.

What works is a frame that is already inside the action, with the subject large enough to read on a phone: at least 35% of the frame height, centred in the middle third where the thumb is not covering it. If the video is a tutorial, the first frame is the finished result, not the ingredients. If it is a story, the first frame is the moment after the thing went wrong.

A practical rule: mute the first frame and look at it as a still image for one second. If you cannot say what is happening and what is at stake, the frame is a title card and it is costing you the view.

Cut cadence: change on the beat, not after it

Cuts land late for a boring reason — the editor drops the cut at the frame where the move looks finished, which is usually 3–6 frames after the beat. On a phone at arm's length, that reads as sluggish even when nobody can name why.

Set the cut on the beat and let the motion finish under the cut. A workable starting cadence for a 24–30 second vertical short:

  • A visual change every 1.2–2.0 seconds in the hook and payoff sections.
  • One longer hold of 3–4 seconds where the explanation is dense; a viewer who is reading needs the frame to stop moving.
  • No two consecutive cuts to the same shot size, and never two hard cuts within 0.4 seconds unless the rhythm is the point.

Beats come from the track, not from feel. Mark the downbeats once on the timeline, then place cuts on downbeats and let the in-between frames carry motion. When a cut has to happen off-beat — to hide a jump in generated footage, or to drop a word — do it on a motion frame so the eye follows the movement instead of the seam.

Caption rhythm is a second edit

Captions in 9:16 are not subtitles for the hearing-impaired; they are the primary read for a large share of viewers watching with sound off. That changes the grammar:

  • Two to four words per line, one line on screen at a time. Four words at 60 px is already the practical limit on a 1080×1920 frame.
  • Each line stays for as long as it takes to read it twice, typically 12–16 frames at 30 fps.
  • Captions sit in the middle-lower band, above the platform's own UI: keep everything inside a 120 px margin from the bottom and 60 px from the sides, because the caption bar, the username and the action buttons all live there.
  • Contrast is a rendering decision, not a taste one: white text with a 2 px dark outline or a solid 80% opacity plate.
  • Do not caption what the frame already shows. "She opens the box" under a shot of a box being opened is two signals competing for the same attention.

The rhythm to aim for is call and response between voice and text: the caption leads the word by 2–3 frames so the eye arrives just before the ear. That tiny offset is what makes a fast vertical edit feel readable rather than frantic.

The loop point is where the rewatch comes from

Shorts get watched more than once when the end and the start fit together. The mechanics are simple and almost always skipped:

  1. Match the composition of the last frame to the first frame — same subject position, same light, within about 2% of the frame.
  2. Make the last spoken line a question the first line already answers, or a clause that only completes on the rewatch.
  3. End 4–8 frames before the action resolves, so the loop feels like a continuation rather than a restart.
  4. Do not add an end card, a subscribe prompt or a fade to black in the final second. Each of them breaks the seam and tells the viewer the video is over.

Check it by playing the last two seconds and the first two seconds back to back, ten times. If the loop is audible — a breath, a room tone cut, a music downbeat restarting — the viewer hears it too, and the illusion of one continuous thing collapses.

What to measure, and what to change first

Retention graphs on vertical platforms show three readable points: the drop inside the first two seconds, the hold through the middle, and whether the tail rises or falls. Read them in that order, and change one variable per batch of three videos, never per video.

Symptom on the graphFirst thing to change
Steep fall before 2 sFirst frame — subject size, motion, or the six-word promise
Fall between 3 and 8 sDelayed proof; move the result earlier
Flat middle, weak tailCut cadence too even; add one long hold and one hard cut
Strong tail, low total reachHook is working, packaging is not — thumbnail frame and first caption
Good reach, low rewatchLoop point; check the last frame against the first

Before you publish

  • First frame readable as a still, subject 35%+ of frame height
  • No logo, fade or black frame anywhere in the first second
  • One promise, six words or fewer, spoken and captioned together
  • Every caption line 2–4 words, 12–16 frames, inside the safe band
  • Cuts on downbeats, no two consecutive same-size shots
  • Last frame matches the first composition; no end card in the final second
  • Loop watched ten times back to back with sound on

None of this makes a weak idea strong. It makes a strong idea survive the first second and get watched twice, which is the only thing the algorithm is actually measuring.

Toolbox for this workflow

  • GammaTurns a shot list or a client brief into a presentable deck without opening a slide app. Useful when the storyboard has to be shown to someone else.

Some links in this box are affiliate links. If you subscribe through one, Omong Studio may earn a commission at no extra cost to you. These are tools used in the workflow above; nothing here is paid placement.

Kits from the studio

Independent kits from the studio: prompt packs and templates that pair with the workflows in these posts.

Work with the studio

Want this made for your brand instead of by you?

Omong Studio takes commissions for long-form storytelling, 9:16 short-form ads and original brand songs. Pilots from $95, 2–3 business days, and a fixed quote in writing before any work starts.

See what a commission includes →

All posts