shortshort
GUIDE

How do you reframe a horizontal 16:9 video to vertical 9:16?

Turn a horizontal 16:9 talk, podcast or webinar into vertical 9:16 clips without losing the speaker, slides or group shots, and meet YouTube Shorts rules.

The short answer

Crop a full-height 9:16 window out of the 16:9 frame and move it with the speaker's face, cutting instead of panning at camera changes. Show slides, charts and group shots whole rather than cropping them. That window keeps 31.6% of the width, so a 1080p source is scaled up 1.78x. A 16:9 upload stays a regular video: YouTube Shorts are square or vertical (checked September 2026).

Sources checked on 22 September 2026, each listed at the end of this guide with its date.

Why a 16:9 recording has to be cropped to become a Short

YouTube counts a video as a Short when it has a square or vertical aspect ratio and lasts up to three minutes. The rule applies to videos uploaded on or after October 15, 2024. YouTube tells creators who want to avoid Shorts classification to use a wider ratio such as 16:9, so a horizontal 16:9 upload stays a regular video even when it is under three minutes.

Placing the whole 16:9 frame inside a vertical canvas with black bars for an entire clip is not a good fix. YouTube's general advice is to avoid adding padding or black bars directly to the video, because padding interferes with the player's ability to adapt to the video size and the viewer's device. On a computer, YouTube's standard ratio is 16:9, and it may add its own padding around a 9:16 video in the browser.

How much of a 16:9 frame a 9:16 crop keeps

A 9:16 window cut at full height from a 16:9 frame keeps 31.6% of the width. The rest is outside the crop, so the real question is where to place that slice at each moment. At full height (x1.0), the slice can only move left and right. Zooming in shrinks it further, and each source pixel is enlarged more. In a shortshort 9:16 export, a push-in on a tracked face peaks near x1.11 (see the next section), and the hard ceiling of x1.6 is reached only when a passage is not tracked.

Source resolution affects how sharp the result looks. From a 1080p recording, the slice is scaled up to reach 1080 x 1920. For a 4:5 export, the window at full height keeps 45% of the width instead, 864 x 1080 px from a 1080p source, scaled up 1.25x to 1080 x 1350. The 4:5 shot zoom goes to x1.4 at most, to bring a small face to about 20% of the frame height: a 617 x 771 px crop, scaled up 1.75x, about as sharp as the 9:16 slice at x1.0. From a 3840 x 2160 recording at x1.0, the crop is scaled down instead. A YouTube link is imported at up to 720p, and at 480p or 360p when the 720p file would exceed 2 GB, so for the sharpest crop upload the original file rather than the link.

SourceZoomCrop taken from the sourceTo reach 1080 x 1920
1920 x 1080x1.0608 x 1080 pxscaled up 1.78x
1920 x 1080x1.11 (push-in peak)546 x 971 pxscaled up 1.98x
1920 x 1080x1.6380 x 675 pxscaled up 2.84x
3840 x 2160x1.01215 x 2160 pxscaled down
1280 x 720 (YouTube link import)x1.0405 x 720 pxscaled up 2.67x

Keeping the speaker in frame

A good vertical crop follows the face, keeps the whole face inside the frame and avoids constant zooming. In shortshort, the tracker samples each selected passage 3 times per second with MediaPipe Face Detector. It aims to leave 0.35 face-width of space on each side of the face, and narrows that space rather than cut the face. In an extreme close-up, when the face is wider than the frame itself, the crop keeps the middle of the face, which fills 90% of the frame width, rather than switching to the whole picture. It sets one base zoom level per shot: in 9:16, x1.04 when the face is 28% of the frame height or less, x1.0 otherwise; in 4:5, the level that brings a small face to about 20% of the frame height, at most x1.4. This base zoom changes only at a camera cut or a switch to another face, so the frame does not pump.

At a camera cut the tracker detects, the crop never pans across it. It holds its position, then jumps to the new framing, and the cut is located to within 1/30 s, one frame of the 30 fps export. Within a shot, the frame stays still through small moves of the speaker and the jitter of the detector, and follows larger moves smoothly; a jump of more than about a third of the frame width within 0.4 s is played as a cut instead of a fast pan. If the detector loses the face for up to 1.5 s and finds it again in about the same place, the gap is bridged, but never across a cut.

A few words the AI marks as punch words get a gentle push-in centred on the face: it eases in over 600 ms, holds for at least 500 ms and eases out over 900 ms, with up to 7% more zoom on a face under 30% of the frame height, 4% on a larger one, and less when a tighter crop would cut the face. It happens only on a tracked face, never on a full-image shot, during a pan or just before a cut, and in 9:16 tops out near x1.11 (the x1.04 base times 1.07); in 4:5, near x1.5. In a 9:16 export, a source that is already 9:16 keeps its whole frame, with no zoom and no panning.

Slides, charts and wide shots: show them whole or crop around the subject

A slide cut down to a third of its width usually loses its text. The fix is to show the full picture for those moments only. In shortshort, the AI analysis flags moments to keep whole: slides, charts, essential on-screen text, or several subjects a vertical crop would cut off. Those moments are shown whole across the 1080 px width (1080 x 608 for a 16:9 source), over a blurred, darkened copy of the same video rather than black bars, even when a face is present.

Other wide shots are not shown whole. A wide shot with no usable face that the analysis did not flag fills the 9:16 frame, cropped around the subject the analysis pointed to, with a zoom between x1.0 and x1.08. A face under 10.5% of the frame height, which can be the case for a speaker filmed from far away on a stage, gets the same treatment. It is not zoomed in close.

A full-picture moment is still a picture framed inside a taller frame. Keep it for the moments where a crop would lose information, and let the speaker fill the frame the rest of the time.

Two-person podcasts, fast cuts and failed tracking

With two people in one fixed shot, shortshort follows the most prominent face, ranked on the detector's confidence and the face's size. It switches by visible face, not by voice: there is no speaker diarisation. If the other face becomes the most prominent for less than 1.5 s, the frame does not move. If it stays the most prominent for longer, the crop usually switches to it with a clean cut, not a pan. Who is speaking plays no part, so in a steady two-shot the crop can stay on the listener. If two faces carry similar weight and neither is above 20% of the frame height, no face is picked for that moment, and the frame stays where it was on the last face of that shot.

Very fast editing is handled shot by shot. When a passage has more cuts than its 58 framing keys can follow, the shortest shots each get one fixed frame on their own face while the longer shots keep their tracking; beyond 58 shots, neighbouring shots that frame the same place share one frame. If in-browser tracking fails, the short is not tracked and the studio shows a notice saying so.

Vertical specs for YouTube Shorts, Instagram Reels and TikTok

The three platforms do not publish the same kind of figures. YouTube gives rules for organic Shorts. For Instagram Reels, the first row comes from the Instagram Help Center, for organic reels, and the second from Meta's ads guide, for Reels ads. For TikTok, the length comes from its Help Center and from its developer Media Transfer Guide, which covers apps posting through the Content Posting API; the other figures come from its In-Feed ads spec and that developer guide. All were checked on September 22, 2026.

Safe zones matter for captions and slide text. For Reels ads, Meta suggests keeping at least 14% of the top, 35% of the bottom and 6% of each side free of text, logos and other important elements. TikTok's In-Feed ads page gives no single set of margins: the safe zone depends on the dimension, the ad caption length and any extra formats, and TikTok provides downloadable safe-zone files.

In a 9:16 short, shortshort places captions in a fixed block 490 px above the bottom edge of the 1920 px frame, about 25%, above the bottom 484 px that third-party 2026 safe-zone templates give TikTok's interface. That is still inside the bottom 35% Meta suggests keeping clear for Reels ads, and the position cannot be moved. For an ad, turn captions off or use the caption-free copy in the Download all ZIP.

PlatformAspect ratioLengthOther published figuresSource
YouTube ShortsSquare or vertical. YouTube advises a wider ratio such as 16:9 to keep a video out of Shorts.Up to 3 minutes (uploads on or after October 15, 2024)On computer browsers, YouTube may add padding around 9:16 videosYouTube Help
Instagram Reels (Help Center)1.91:1 to 9:16 accepted; 9:16 is the tallestUp to 20 minutes; over 3 minutes not recommended to new audiencesAt least 30 fps and 720 px. No maximum file size statedInstagram Help Center
Instagram Reels (ads spec)9:160 seconds to 15 minutes (Reels ads)1440 x 2560 px recommended for Reels ads. Keep 14% top, 35% bottom and 6% each side clear.Facebook Ads Guide
TikTok9:16 recommended for In-Feed ads, at least 540 x 960 px. 16:9 and 1:1 also accepted for ads.Uploads up to 60 minutes (Help Center, English); apps posting through TikTok's Content Posting API: 3 minutes for all creators, 5 or 10 for someDeveloper guide: 360 to 4096 px per side, 23 to 60 FPS, MP4 and H.264 recommendedTikTok Help Center, TikTok Ads Manager, TikTok for Developers

With shortshort

shortshort reframes each short automatically from one long video: an uploaded file (MP4, MOV or WebM, up to 2 GB) or the link of your own public or unlisted YouTube video, imported at up to 720p. It tracks the face with MediaPipe, keeps one base zoom level per shot, cuts instead of panning at camera changes, and shows the slides, charts and group shots its analysis flags whole, over a blurred copy of the video. Tracking runs on every plan, including Free. In 9:16, a clip added with 'Add a clip' is not tracked: it keeps a fixed crop on the centre of the picture at x1.0, and extending a clip holds the framing of its first or last tracked moment over the added seconds. A 4:5 export tracks every clip again, those included. The crop cannot be moved by hand, so if the speaker sits off-centre, change the clip bounds or pick another passage. Export is MP4 H.264 at 30 fps: 9:16 at 1080 x 1920, or 4:5 at 1080 x 1350 for the Instagram feed, where the face is tracked again for the 4:5 frame rather than the 9:16 crop made wider.

Questions people ask

Can I upload a 16:9 clip as a YouTube Short?

No. YouTube categorizes an upload as a Short when it has a square or vertical aspect ratio and lasts up to three minutes. To keep a video out of Shorts, YouTube tells creators to use a wider ratio such as 16:9, so a 16:9 clip stays a regular video even under three minutes (checked September 2026).

What does shortshort do with slides during a talk?

When its AI analysis flags slides, charts or essential on-screen text, shortshort shows that moment as the full picture across the 1080 px width, over a blurred, darkened copy of the video. Wide shots it does not flag are cropped to 9:16 around the subject instead.

Does shortshort follow whoever is talking in a two-person podcast?

No. It follows the most prominent visible face, not the voice, because it has no speaker diarisation. If the other face becomes the most prominent for 1.5 s or more, the crop usually cuts to it; a shorter change never moves the frame. In a steady two-shot, the crop can stay on the listener.

Will a speaker far away on a stage be zoomed in?

In 9:16, not much: a tracked face stays at x1.0 or x1.04 for each shot. In 4:5, the shot zoom brings a small face to about 20% of the frame height, at most x1.4. Either way, shortshort tracks a face only if it is at least 10.5% of the frame height; a smaller face gets a crop around the subject the analysis pointed to, with a zoom between x1.0 and x1.08.

Sources

The official pages this guide relies on, with the day each one was read.

Related

All 16 guides

Cut your own long video into shorts.

30 credits are free at sign-up, no card. One credit is one minute of source video, whatever the number of shorts.