There are a hundred aspect ratio cheat sheets online and most of them are copying each other. When you check the numbers against the platforms' own documentation, two things become obvious: the ratio and duration specs are mostly reliable, and the safe zone figures, the part that actually decides where you put a face or a caption, are almost entirely invented.

1080x192016:91:14:5CENTRE-SAFE9:16
The mechanical crop is not the bottleneck. Composition discipline at generation time is.

The specs that are actually published

These come from the platforms themselves.

  • TikTok: 9:16 recommended. Minimum 540x960 for vertical, 960x540 horizontal, 640x640 square. Ad uploads up to 10 minutes, 500MB.
  • YouTube long-form: 16:9, up to 7680x4320. H.264 High Profile, closed GOP, 4:2:0. Encode at the frame rate you recorded.
  • YouTube Shorts: 9:16, maximum upload resolution 1080p, up to 3 minutes. For Shorts ads, only the first 60 seconds plays in the feed.
  • LinkedIn: 1:1, 4:5, 9:16 and 16:9 all supported, 3 seconds to 30 minutes, 500MB. It publishes an explicit 5% aspect ratio tolerance, which is unusually precise and useful.
  • X: accepts ratios only between 1:2.39 and 2.39:1 inclusive. Max 40fps, 25Mbps. For ads the file limit is 1GB but the platform's own advice is to stay under 30MB.
  • Pinterest: 9:16 full-bleed at 1080x1920, organic video up to 5 minutes.
  • Threads: accepts 0.01:1 to 10:1, which is the most permissive window anywhere, while recommending 9:16 to avoid cropping.
  • Instagram Reels: 9:16, and the limit went from 90 seconds to 3 minutes in January 2025.

The safe zone problem

A safe zone is the part of the frame the interface will not cover with captions, usernames, buttons and progress bars. It is the single most useful number in social video production, and almost nobody publishes it.

Pinterest is one of only two platforms that does. Its published figures are top 270px, left 65px, right 195px, bottom 790px. On a 1080x1920 canvas that leaves a safe box of roughly 820x860, about 34% of the frame area. Snapchat maintains a safe zone document too.

Pinterest’s published safe zone

Margin reserved on a 1080x1920 canvas, px. The safe box left is roughly 820x860.

0200400600800BottomBottom: 790 pxBottom: 790 px790 pxTopTop: 270 pxTop: 270 px270 pxRightRight: 195 pxRight: 195 px195 pxLeftLeft: 65 pxLeft: 65 px65 px
Pinterest’s published safe zone
EdgeReserved margin
Bottom790 px
Top270 px
Right195 px
Left65 px
Source: Pinterest, Pin specs including published safe zone pixel figures.

TikTok explicitly refuses to publish numbers, and the reason is the important part. From its own documentation: the safe zone size is determined by the ad dimension format and the ad caption length. A longer caption wraps to more lines and pushes the bottom boundary higher. It is a variable, not a constant, which is why TikTok ships template files instead of figures.

One sentence in TikTok's own docs invalidates every fixed-pixel TikTok safe zone guide in circulation.

And those guides disagree wildly. Comparing four current third-party sources on a 1080x1920 frame: TikTok's bottom reservation is given as 320px, 480px and 250 to 300px. YouTube Shorts' top reservation is given as roughly 170px by one and 380px by another, a 2.2x spread. None of them cites a platform document.

What to do instead

Three things that are defensible.

  • The right-hand column is the one reliable figure. Roughly 120px is where three independent sources agree, because TikTok, Reels and Shorts all put engagement buttons there.
  • Compose for a centre-safe box of about 1080x1420, which is roughly 250px off the top and bottom, about 74% of frame height. It is a working heuristic, not a spec, and say so.
  • Check on a real device with a long caption. Since the bottom boundary moves with caption length, the only honest test is the worst case: your longest caption, on the phone, before you publish.

What happens when the ratio is wrong

The behaviour differs by platform and most guides flatten it to "it gets cropped." It does not.

  • YouTube pads, never crops, and it pads white by default, dark grey in dark theme. It explicitly tells you not to add bars yourself. So baked-in black bars sit inside a white frame on a light-theme desktop and look broken. This is the single most common self-inflicted error in the list.
  • YouTube Shorts ads blur. Horizontal assets serve with blurred top and bottom, not letterboxed.
  • X hard-rejects anything outside 1:2.39 to 2.39:1.
  • LinkedIn fails validation outside its 5% tolerance rather than reframing.
  • Instagram crops the grid thumbnail to 3:4 regardless, including Reel covers. Your cover frame needs a 3:4 centre-safe composition on top of its 9:16 one.

Does vertical actually perform better?

Weakly supported, and worth being honest about. There is one genuinely peer-reviewed study: Mulier, Slabbinck and Vermeir in the Journal of Interactive Marketing (2021), three studies including a large field test, finding vertical increased interest and engagement through processing fluency, moderated by age.

Three caveats belong with it. The proposed mechanism is that vertical saves you rotating the phone, which is largely obsolete now that every vertical feed pads horizontal content in place. It measures interest, not revenue. And it compares vertical against horizontal as ad creative, not vertical-native against reframed-from-horizontal, which is the decision you actually face.

The most-cited alternative is a 2019 test by two vendors who sell vertical video tooling, comparing vertical against square rather than horizontal, with no sample sizes disclosed. Everything else is listicles recycling each other.

The defensible conclusion: vertical wins on vertical-first surfaces because the platform gives vertical the full screen and pads everything else. That is a distribution artefact, not proof that vertical framing is inherently more persuasive. Anyone quoting a precise percentage lift is almost certainly quoting 2019 vendor marketing.

The constraint nobody mentions

If you are generating video with AI, the one-master-versus-native question is already decided for you. Veo 3.1 supports exactly two aspect ratios: 16:9 and 9:16. Sora 2 documents 1280x720, 1920x1080 and 1080x1920, which is the same two. Neither can natively generate 4:5, 1:1 or 2:3.

So anyone producing AI video for Instagram Feed, LinkedIn square or Pinterest is already reframing from a 9:16 or 16:9 master, whether they think of it that way or not.

The workflow that follows

  • Generate 9:16 as the master for TikTok, Reels, Shorts, Stories, Snapchat and Threads. Six placements, one master, natively supported.
  • Centre-crop to 4:5 and 1:1 for Instagram Feed, Facebook Feed and LinkedIn. You are removing top and bottom, which is exactly where the safe zones told you not to put anything. Compose for the centre-safe box and the crop is free.
  • Do not reframe 9:16 to 16:9 for YouTube long-form. That crop throws away most of the frame with no safe-zone discipline protecting it. Generate a separate 16:9 master, which both models support natively.
  • Never bake letterbox bars into the master. See the white padding problem above.

Two native masters, every placement

Veo 3.1 and Sora 2 generate only 16:9 and 9:16 natively

9:16 master9:16 masterTikTok, Reels, Shorts, Stories,Snapchat, ThreadsSix placements, one mastercentre-crop4:5 and 1:14:5 and 1:1Instagram Feed, Facebook Feed,LinkedInSeparate 16:9 masterSeparate 16:9 masterYouTube long-formDo not reframe 9:16 to 16:9
Compose for the centre-safe box and the crop is free.

Auto-reframe with subject tracking is commodity now, so the mechanical crop is not the bottleneck. Composition discipline at generation time is. One last honesty note: I found no credible measured study comparing reframed-from-master against natively-produced-per-platform creative. That recommendation is reasoning from constraints, not evidence.

Sources

Gabriel Brien

Gabriel Brien

Founder of Crimson Spark Agency. AI filmmaker and creative technologist, writing from running this work daily.

More about Gabriel →