Aspect Ratios and Safe Zones

Published October 2026 · 10 min read

Most people decide the aspect ratio at export. In AI video, that is too late. The frame shape is baked into every generation, and converting afterwards means either cropping away composition you carefully built or upscaling empty space that was never there.

The rule is simple and worth internalising: generate in the ratio you will deliver. Everything below explains why, and what else changes when the frame is vertical.

What each platform actually wants

These are the current delivery specifications. Platforms adjust them, so verify before a large job, but the ratios have been stable for years.

PlacementRatioTypical resolution
YouTube (standard)16:91920×1080
YouTube Shorts9:161080×1920
TikTok9:161080×1920
Instagram Reels9:161080×1920
Instagram feed4:51080×1350
Instagram square1:11080×1080
LinkedIn feed1:1 or 4:51080×1080 / 1080×1350
Website hero16:9 or 21:91920×1080 / 2560×1080

Note that 9:16 dominates short-form, and that 4:5 occupies more vertical screen space than 1:1 on Instagram without being cropped by the interface. If you are producing one asset for multiple placements, the question is not which ratio to pick but which one to protect.

Safe zones: the part of the frame you cannot use

On vertical platforms, a substantial portion of the frame is covered by interface elements: captions at the bottom, navigation at the top, action buttons on the right. Anything important placed there is invisible to the viewer.

On a 1080×1920 canvas, the approximate reserved areas are:

That leaves a practical safe rectangle of roughly 930×1400px, offset slightly left of centre. Everything that matters — faces, product, text, logo — should sit inside it.

These numbers are estimates and they do change. The safe approach is to keep critical content within the central 85% and treat the outer edges as disposable. If you want the exact figures for your current deliverable, the aspect ratio tool converts between ratios and shows the resulting dimensions.

Vertical is not horizontal rotated

The most common compositional error in AI video is generating a horizontal composition and cropping it to vertical. The result looks cropped because it is: you lose 60% of the frame width and keep whatever happened to be in the middle.

Vertical framing has different rules:

When writing the prompt, the camera slot should reflect this. “Wide establishing shot” is a horizontal instinct. For vertical, “full body, subject centred, camera slightly below eye level looking up” uses the frame properly.

Protecting a frame across multiple ratios

If one asset must work at 16:9, 9:16 and 1:1, you cannot compose optimally for all three. The practical solution is a protected centre: compose so that the central square (1:1) contains everything essential, and treat the wider 16:9 area as extension.

The reverse — generating vertical and extending to horizontal — does not work, because there is no image information outside the vertical frame to extend into. Generation is the only place the extra width can come from, so generate wide and crop narrow, never the other way around.

This is the one case where generating in a ratio other than the deliverable makes sense, and it only works in one direction.

How much you actually lose when you crop

“Crop it later” sounds free. It is not, and the numbers make the cost obvious. Starting from a 1920×1080 frame:

Target ratioResulting sizeWidth retainedUsable?
16:9 (native)1920×1080100%Yes
4:5864×108045%Marginal
1:11080×108056%Marginal
9:16608×108032%No

The vertical crop keeps less than a third of your frame width and produces a 608px-wide image, which is below every platform's expected delivery width. Any composition built across the 16:9 frame is destroyed, and upscaling 608 to 1080 does not recover detail that was never captured.

This single table is the strongest argument for deciding the ratio before you generate. It is also why the “generate wide, crop narrow” advice has a floor: it works down to about 4:5 and fails completely at 9:16.

The resolution trap

Generating at 1080p and delivering at 1080p sounds correct until you crop. A 1080p 16:9 frame cropped to 9:16 yields a 608×1080 image — well below the platform's expected width, and visibly soft after upscaling.

Two ways out. Generate natively in 9:16 so no crop is needed, or generate at a higher resolution than you deliver so the crop still lands above 1080 wide. The second option costs more per second, which is why the first is usually correct.

Worth noting: on most models, 9:16 generation costs the same per second as 16:9, because you are paying for duration rather than pixel count. There is no financial reason to generate horizontal and crop.

Four mistakes worth avoiding

Deciding the ratio at export. The single most expensive habit. Every generation before that decision may be unusable.

Placing text in the interface zone. Captions that sit under the platform's own caption bar. Keep text above the bottom 400px.

Assuming the platform will letterbox. Vertical placements do not letterbox — they crop or they show your video in a small box with blurred bars, which reads as low effort.

Forgetting that the same video goes to multiple places. A 16:9 export posted to Reels will be cropped by the platform automatically, at a moment and in a position you do not control. Always crop deliberately in the edit.

How to check your own safe zones

Published safe-zone diagrams go out of date, and they differ between the preview in a desktop browser and the actual app. The reliable method is empirical, and it takes five minutes.

  1. Export a test clip at your delivery resolution with a visible grid overlaid — thin lines every 100px, numbered.
  2. Post it as a private or draft post on the target platform.
  3. Open it in the mobile app, not the desktop preview, and screenshot the playback screen.
  4. Measure where the interface elements actually fall on your screenshot.

Repeat once every few months, or whenever a platform announces a redesign. The numbers move more often than the ratios do, and a layout change that shifts the caption bar upward by 60px is enough to bury a line of text you carefully placed.

If you produce regularly for one platform, keep the measured safe rectangle as a reference layer in your editor. Compositing against it takes no extra time and eliminates an entire category of rework.

Text and captions specifically

Text is the element most often ruined by safe zones, because text is the one thing that cannot be partially covered without becoming unreadable.

Three practical rules. Keep all text within the central safe rectangle, never in the outer 15%. Size text for a phone held at arm's length, not for your monitor — a 24px caption that looks generous on a 27-inch display is illegible on a phone. And leave a margin below any text block, because the platform's own caption bar can expand when a user enables captions.

For AI-generated video there is an additional constraint: do not generate text into the frame at all. Model-rendered text is unreliable, covered in the nine tells article, and even when it works it is fixed in position. Add text in the edit, where you can move it inside the safe zone after the fact.

A note on 21:9

Ultrawide is tempting for website heroes and looks impressive on a large monitor, but it is a poor default for social delivery. No major platform displays 21:9 natively, so the frame will be letterboxed or cropped by the player, and the extra width you paid to generate is thrown away. If you want the look, compose at 16:9 and add the bars in the edit — that way the bars are removable and the underlying frame still works everywhere else.

A short workflow

  1. List every placement the asset will appear in before generating anything.
  2. If all placements are vertical, generate 9:16 and compose into the frame.
  3. If placements are mixed, generate 16:9 with everything essential inside the central square, then crop deliberately per platform.
  4. Keep faces, product and text inside the central 85% of the frame.
  5. Check the final export at phone size, not on your monitor. Interface overlap only becomes obvious at actual size.

Related: the video specs calculator converts between ratios and resolutions, the file size calculator estimates the deliverable weight before you export, and the shorts versus long-form comparison covers the editorial side of choosing a format.