Aspect ratio

The arithmetic

A ratio is written two ways and they mean the same thing. 16:9 is the colon form — divide it out and you get 1.78, the decimal form, which is what you compare against a platform's accepted range. 9:16 is 0.5625, 4:5 is 0.80, 1:1 is 1.00, 4:3 is 1.33. Cinema tends to use the decimal form with a trailing :1 — 1.85:1 for flat, 2.39:1 for scope — because those numbers do not reduce to small whole numbers.

Ratio is not resolution. 1280x720, 1920x1080 and 3840x2160 differ by a factor of nine in pixel count and are all 16:9. Resolution controls how finely the picture is sampled; ratio controls what is inside the frame. Conflating them is the single most common error in this subject, and it survives because delivery presets bundle the two together until you meet a file where they come apart.

Which they do, because pixels are not always square. Three quantities are involved, related by one identity:

SAR (stored pixels, w:h) × PAR (shape of one pixel) = DAR (what you see)

In a modern square-pixel file PAR is 1 and the storage ratio equals the display ratio, which is why the distinction feels academic. It stops being academic the moment you open older footage. An HDV file stores 1440x1080 — a 4:3 grid of pixels — with a PAR of 1.333, and displays as 16:9. PAL DVD stores 720x576 with a PAR of 64:45; 1.25 × 1.422 = 1.78, so it too displays 16:9. This is anamorphic storage: the picture is squeezed on the way in and stretched on the way out. Software that reads the pixel dimensions and ignores the PAR metadata is the reason people occasionally get a stretched or squashed export from footage that previewed correctly.

16:9 itself is a compromise, deliberately. In 1984 Kerns H. Powers, working on SMPTE's high-definition group, cut equal-area rectangles for the five ratios then in use — 1.33, 1.66, 1.85, 2.20 and 2.35 — and overlapped them at their centres. Every rectangle fitted inside a common outer rectangle of roughly 1.78, and every rectangle contained a common inner rectangle of the same 1.78. It is close to the geometric mean of 4:3 and scope: √(1.33 × 2.39) ≈ 1.78. The ITU adopted it as the international widescreen standard around 1990. Nobody shoots 16:9 because it is beautiful; it is the shape that wastes the least when everything else has to be fitted into it.

The ratios that matter, and what each is for

A note on the cinema row: the scope ratio is 2.39:1, not 2.35:1. The 35mm anamorphic frame was 2.35 until SMPTE revised it in 1970 — the image height was trimmed slightly so that negative-assembly splices stopped flashing at the top of the frame — and it has been 2.39 ever since. "2.35" survives as a name for the look rather than as a measurement.

Platforms accept a range and reward a point inside it. X accepts anything from 1:2.39 to 2.39:1 but renders 16:9, 1:1 and 9:16 cleanly. LinkedIn accepts 9:16 through 16:9. YouTube's player adapts to whatever you upload by adding bars, and routes square-or-taller uploads under the Shorts length limit (3 minutes since October 2024) into the Shorts player rather than the main one — so on YouTube the ratio is not only a framing choice, it selects which product your video lands in.

Safe areas: the ratio is not the whole frame

Your 1080x1920 vertical video is never watched at 1080x1920. The app draws its own interface on top: a username, a caption, a sound label, an action rail, a follow button. The safe area is what is left. Approximate current measurements, all inside the same 1080x1920 canvas:

• TikTok — top band about 150-200px (username, sound), right rail about 120-160px wide (like, comment, share), bottom about 250-480px depending on how long the creator's caption is. The right rail is the one landscape editors forget, because nothing in a 16:9 workflow ever covers a vertical strip of the picture.
• Instagram Reels and Stories — Meta has been consolidating these toward one shared 9:16 safe zone. Keep hard-coded text clear of roughly the top 250px and bottom 450px, leaving a central band about 1220px tall.
• YouTube Shorts — roughly the top 180px (title, status bar) and bottom 390px (channel info, call-to-action). Guidance commonly given is a centred 900x1160 region for critical text.

A centred region about 900px wide by 1150px tall clears the top and bottom furniture on all three. TikTok's right-hand action rail is the extra constraint that band does not clear on its own — staying inside a centred column about 760px wide keeps text off it too. Two honest caveats: these numbers are measurements of shipping app interfaces, not published standards, so they change without notice and differ between devices and between a platform's organic feed and its ad placements. And the practical consequence is asymmetric — the same edit can be perfectly composed in 16:9 and broken in 9:16 without a single pixel of the picture changing, because what changed is the furniture sitting on top of it.

Reframing is a cropping decision with a tracking problem inside it

Converting a finished 16:9 edit to another ratio is where the abstraction stops being abstract. Start from a 1920x1080 frame and keep the full height. The crop window you are allowed is:

• 9:16 → 608 x 1080 — 31.7% of the width. Roughly two thirds of the picture is gone.
• 4:5 → 864 x 1080 — 45% of the width.
• 1:1 → 1080 x 1080 — 56% of the width.

Padding is not free either. Fit a whole 16:9 shot into a 1080x1920 vertical canvas and the picture occupies 1080x608 — about 32% of the output area. The other 68% is bars, or a blurred copy of your own frame. Nothing adds real picture, because the picture outside the original frame was never recorded.

So every reframe is two separate decisions, and confusing them is what produces bad output.

One: crop or fit? This is a question about content, not taste. Footage with a single centre-weighted subject — a person talking — should be cropped, because a vertical crop of a talking head is what vertical video is. Footage whose content runs to the edges of the frame is a different case entirely: a game recording with a HUD in three corners, a screen recording, a wide establishing scene. Crop that and you have not reframed it, you have truncated it, and the parts you removed were the parts carrying information. Fitting the whole frame in — over black bars or a blurred backdrop — is the honest conversion there, even though it looks like the lazier option.

Two: where does the window sit? If the subject holds still, one horizontal offset solves the whole video. If the subject moves — a presenter who walks, an interview that cuts between two chairs, a handheld shot that drifts — then no single offset is correct throughout, and the crop window has to move with the content. That is the subject-tracking problem, and it is why reframing is properly a per-shot decision rather than a per-file one. A tool that gives you a single number is approximating; a tool that gives you keyframes is asking you to do the tracking by hand.

The scenario where this bites

You have a two-person interview, shot 1920x1080, the guests seated at roughly one third and two thirds across the frame. That is 640px apart. The client wants a 9:16 clip for Reels.

The 9:16 crop window is 608px wide. The two people are 640px apart. No crop position holds both. Centre it and you land between them and cut both at the shoulder. Aim it at one and the other leaves the frame entirely. There is no export setting that fixes this, because it is not an export problem.

The three real answers are all edits. Cut between two separate crops, one per speaker, following the audio — which is an editorial pass that did not exist in the 16:9 version. Stack two crops vertically as a split-screen. Or fit the whole 16:9 shot into the top of the 9:16 canvas and use the remaining 1312px for captions and titles, which is why so much interview content on vertical platforms looks that way — it is not a style, it is the arithmetic. Aspect ratio conversion is an editorial pass wearing an export setting's clothes.

Common mistakes

How this works in Valmera

You ask for a ratio in plain English and the agent converts the output frame. Whole-video targets are 16:9, 9:16, 1:1, 4:5 or the source ratio, in one of three modes: crop, pad (bars) or blurred pad. There is a separate operation for shifting the visible frame mid-video, which additionally allows 4:3.

The automatic reframe makes both decisions above by measuring, not assuming. It samples five frames spread across the kept footage — the edit, not the raw upload, so framing follows what the viewer will actually see — and detects faces in those pixels to aim the crop. When no face is found it asks the prior question instead: it integrates the picture's detail over the exact window the renderer would take, and if less than 55% of that detail survives the crop it fits the whole frame in over a blurred backdrop rather than cropping, and tells you it did that and why. That is the difference between a talking head, where a crop is the reframe, and a wide game recording or screen capture, where a crop is a truncation.

Two supporting details. Output resolution never exceeds the source's pixel budget: a 1920x1080 source at 9:16 renders 1080x1920, but a 1280x720 source at 9:16 renders 720x1280 rather than being upscaled to look like more. And burned-in captions rescale to the new frame, with bottom-anchored captions in a vertical output held at least 13% of the frame height above the bottom edge — clear of the caption and CTA furniture described above.

What it does not do. The crop focus is a single fixed point for the whole video. It is not per-frame motion tracking, so it will not follow a presenter who walks across the room; when the measured subject drifts across the sampled frames the tool says so and offers the fitted conversion instead, which is a warning rather than a solution. It does not draw platform safe-area guides or overlays. It does not offer arbitrary ratios: the whole-video targets are the ones listed above, so there is no 2.39:1. And it will not generate frame extension — padding is bars or a blurred copy of your own picture, never invented content. Resizing is one of the things the agent does, not the whole of what it does.

Related terms

• Auto-reframe — the automated version of the crop-placement problem described above.
• Shot detection — finding the shot boundaries that make reframing a per-shot decision rather than a per-file one.
• Burned-in captions vs. subtitles — burned text is inside the frame, so it inherits every safe-area constraint.
• Edit decision list — where the output frame is stored as a decision rather than baked into pixels.
• Proxy editing — why the resolution you preview at and the resolution you export at are different numbers.
• B-roll — inserted footage that may not share the main footage's ratio, and has to be fitted or cropped on the way in.
• All glossary terms

Frequently Asked Questions

Reframe by describing the platform

Upload a landscape edit, ask for 9:16, and the agent measures whether a crop or a fit is the honest conversion. 50 free credits, no card.