You cut the master in 16:9 because that is what YouTube wants. Now the same edit has to live on TikTok, Reels, and Shorts, and all three want 9:16. The lazy path is a center crop, and the center crop is where good shots go to die: the subject takes two steps to the left and walks out of frame, the product sits on a rule-of-thirds line and gets amputated, the title card loses half its words. The better path is reframing — recomposing each shot inside the new canvas instead of blindly chopping it — and in 2026 the recomposition can be driven by an AI tracker that follows your subject through the clip. This tutorial walks the full conversion inside Skrrol's browser studio: what each reframe mode actually does, the exact steps, when a letterbox beats a crop, and how to rescue the shots where automation gets it wrong.
Why a straight center crop loses the shot
Start with the arithmetic, because it is worse than most people intuit. A 1920×1080 source converted to 9:16 at full height keeps a window roughly 608 pixels wide. That is less than a third of the horizontal picture — more than two thirds of everything you shot gets thrown away. A crop that severe is survivable only if the thing that matters happens to live inside that middle strip for the entire duration of the clip.
It almost never does. Horizontal footage is composed for horizontal reading: subjects placed on thirds lines, negative space that lets a frame breathe, two-shots with a person on each side, walk-and-talks where the subject drifts across frame as the camera pans. A fixed center window ignores all of that composition and bets everything on the middle 608 pixels. The bet pays off for a locked-off, dead-centered talking head and loses for nearly everything else. The remaining naive options are worse. Scaling the whole frame down to fit the vertical width produces a tiny strip of video floating in empty space. Stretching the image to fill the canvas distorts every face and product in it — which is why stretching is the one option Skrrol's converter refuses to offer. Every conversion is a letterbox, a crop, or a tracked crop, and the proportions of the original pixels are always preserved.
What AI reframing actually does
Reframing turns the crop window from a fixed rectangle into a moving one. Instead of deciding once where the 9:16 window sits, the window's position becomes animated data that can change across the clip, so the conversion follows the action instead of hoping the action stays put.
In Skrrol, the following is done by the same on-device AI model that powers Smart Cutout. The model segments the most prominent subject in the shot — a person, a pet, a product — and the tracker shifts the vertical window so that subject stays centered as they move. Because the tracking is built on per-frame segmentation rather than a one-time detection, it holds onto a speaker pacing across a stage or a pair of hands working along a countertop, and the resulting camera path stays smooth instead of jittering. Everything runs locally in your browser: footage is never uploaded, inference happens on your device, and the export carries no watermark. The same machinery works in both directions — a 9:16 vertical source reframes out to 16:9 just as cleanly — and across every aspect the studio supports: 16:9, 9:16, 1:1, 4:5, 21:9, and custom sizes.
The four reframe modes, and what each is for
Aspect conversion in Skrrol is a per-clip decision, not a per-project one. The project's target aspect sets the canvas; every clip on the timeline then carries its own reframe mode. There are four.
- Letterbox — the full 16:9 frame scaled to fit inside the 9:16 canvas, with black bars above and below. Nothing is cropped, everything is smaller. The zero-loss option.
- Center crop — a fixed full-height window cut from the middle of the frame. Fast and predictable. Right for locked-off shots where the subject genuinely sits in the center.
- Manual — you place the crop window yourself and can keyframe its position over time. Total control, most effort. The mode for precision work and unusual subjects.
- Subject track — the AI follows the person or product and moves the crop to keep them centered for the whole clip. The default for anything with motion in it.
A real conversion mixes modes. A typical YouTube-to-TikTok pass runs subject track on the A-roll where the presenter moves, center crop on locked-off product inserts, and letterbox on the one archival clip where the whole frame is the point. Because the mode lives on the clip, you make each of those calls independently and preview them together on a single timeline.
Step by step: converting 16:9 to 9:16 in Skrrol
The full path from horizontal master to vertical deliverable takes six steps.
- Set the project aspect. Open project settings and switch the target aspect to 9:16. The canvas reshapes to match, and every clip on the timeline now renders inside the vertical frame.
- Choose a reframe mode per clip. Select each clip and pick Letterbox, Center Crop, Manual, or Subject Track. Start with Subject Track for anything containing a person or a moving product.
- Preview the result. Scrub the timeline end to end and watch how each clip lands in the new aspect. Most clips will be fine on the first pass; flag the ones where the framing looks wrong and fix only those.
- Refine the subject path. Where the tracker misses a beat — a fast move, a second person entering, a moment where the wrong thing gets followed — drop manual keyframes on the clip's crop position. Your keyframes override the AI for exactly those frames, and the automatic track resumes on either side.
- Export to the target aspect. The render writes a true vertical file at the aspect's native resolution — 1080×1920, or the 4K equivalent — with portrait metadata baked in so platforms recognize the orientation and display it without rotating.
- Re-export for additional aspects. Duplicate the project, switch the target to 1:1 or 4:5, and render again. The source media never changes — only the framing does — so a square or 4:5 feed cut costs minutes, not another edit.
Two details worth knowing before you render. Audio passes through untouched — aspect conversion reframes the picture only, so dialog, music, and mix are identical across every deliverable. And the vertical export is genuinely vertical: a native 1080×1920 raster, not a horizontal file with bars burned into it, which matters because platforms treat a true portrait file differently from a disguised landscape one.
Letterbox, crop, or subject track: making the call
Subject track earns its place as the default for single-subject motion: walk-and-talks, cooking hands, pets, unboxings, demos — anywhere one clear subject moves against a background. The tracking handles most single-subject shots well, and those shots are the bulk of most timelines.
Center crop wins when the framing is already vertical-friendly. A podcast setup with the host dead center, a tripod talking head, a product on a turntable in the middle of frame — tracking adds nothing there, and a fixed window is perfectly predictable. If nothing moves, do not ask a model to follow it.
Manual is for shots where the subject a model would pick is not the subject you mean. A wide landscape where the point of interest is a small climber on the ridge. A screen recording where the story is the cursor. Keyframe the window where you want it and the shot does exactly what you decided.
Letterbox is for shots where the composition is the content. Establishing wides, archival material, drone reveals — any frame where cutting away two thirds destroys the meaning. It is also the honest answer for full-width on-screen text you cannot rebuild. The cost is real: the image gets smaller inside a feed built for full-bleed vertical, so treat letterbox as an occasional beat inside a vertical edit, not the whole video. And when two subjects sit on opposite thirds of the frame, accept that no single window holds both — letterbox the two-shot, or use manual keyframes to move the framing between them as each one speaks.
Common failure cases and how to fix them
Automated reframing fails in predictable places, and every failure has a cheap fix.
The subject leaves the window. Fast lateral movement, a subject who exits and re-enters, or attention cutting to a second speaker can outrun any smooth camera path. Scrub the preview, find the frames where the framing drifts, and drop manual keyframes on the crop position there. The keyframes override the AI exactly where you place them and hand control back afterward — you are correcting seconds, not redoing the clip.
The tracker follows the wrong thing. The model isolates the most prominent subject in frame, and in a crowded shot the most prominent subject is not always your subject — a passerby crossing near the camera can steal the window for a beat. The fix is the same manual-keyframe override through the contested frames. For genuinely multi-subject scenes, plan on manual keyframes from the start; automated tracking is built around a single dominant subject.
On-screen text near the edges. A lower-third or title built for the full 16:9 width cannot survive a crop to the middle third — you keep half a headline. Three fixes, in order of preference. If you still have the project, strip the baked-in title and set it again inside the vertical canvas after conversion; titles belong to the deliverable, not the master. If the text is burned into the source, switch that clip to letterbox so the full line survives. If only part of the text matters, use manual mode to frame the crop around the words that count. And when you do re-set text for 9:16, keep it in the middle band of the frame — vertical platforms lay interface elements over the bottom and the right edge, and captions parked there get covered.
The shot is too chaotic for any window. Handheld footage where camera and subject both move, or action spread across the full width of the frame, sometimes cannot be recomposed into a third of its canvas and still read. That is not a tracking failure — it is a composition fact. Letterbox the shot, shorten it, or cut around it. A vertical edit does not owe every horizontal shot a place.
Reframe or cut out? When Smart Cutout is the better tool
Reframing keeps the original background and moves a window across it. Sometimes the stronger vertical treatment is to remove the background entirely — lift the speaker out of the wide office shot and place them over motion graphics, stacked B-roll, or a clean backdrop built natively at 9:16. That is a job for Smart Cutout, the studio's AI segmentation tool, which runs on the same underlying on-device model that drives the reframe tracker.
Smart Cutout produces a per-frame alpha matte of the most prominent subject with no green screen required. Edge refinement keeps hair from collapsing into a cardboard outline, temporal smoothing suppresses frame-to-frame flicker, and a manual brush lets you paint corrections on any frame that then propagate to the frames around it. Like the converter, it runs entirely on your device — nothing uploads. The cutout-style vertical workflow: run Smart Cutout on the source clip, build a 9:16 background on the track below, and composite the isolated subject on top at whatever scale the vertical frame wants. It is more work than a tracked reframe, and it is the right call when the original background would waste the canvas — a cluttered office adds nothing to a tutorial, but a full-height presenter over purpose-built graphics fills the frame with signal.
One master, every platform
The conversion workflow is not really about 9:16. It is about ending the era of editing the same video twice. Master once in 16:9 at the best quality you have, then derive: duplicate the project, switch the aspect, adjust the handful of clips that need a different mode, and export. A 9:16 cut for TikTok, Reels, and Shorts. A 1:1 or 4:5 cut for feed placements. Even a 21:9 pass if a cinematic trailer format is on the docket. The source media sits untouched through all of it; only the framing decisions change per deliverable.
If you shoot with conversion in mind, the vertical pass gets faster with every project: keep the subject near the center of frame, keep critical action away from the left and right edges, and leave titles off the master so each aspect gets text set for its own canvas. One edit, every platform — and no version of your video where the subject is standing just out of frame.