Safe zones are not guidelines, they are constraints

01Where does platform UI actually cover a vertical ad?

Along the top, along the bottom, and down one side. A bar or a gradient across the top carrying the account row. A column of action controls down one edge - like, comment, share, the audio disc. A block along the bottom holding the handle, the caption, the call-to-action button and, on some surfaces, a progress bar. The picture keeps playing underneath all of it. Your composition does not.

The same frame, with the furniture on it

Tap the marks

1234

Our own frame, from a fitness studio demo. Note where the wordmark sits. It reads perfectly in this still and it is the first thing an account row would land on.

The expensive failure is a process that treats the canvas as flat and the furniture as an afterthought, so the problem only surfaces at the end, where fixing it costs the most.

02Why this post will not print the pixel numbers

Because they move, and a wrong number in a blog post outlives the spec that made it wrong. Meta publishes its own safe-zone guidance for Stories and Reels ads, and it publishes per-placement specs in its ads guide. I opened both on 1 September 2026 to make sure they existed and to link them properly, and I am deliberately not copying their figures into this page. In six months somebody will find this post through a search, trust the number, and build to it.

03Design to the smallest guaranteed rectangle, not to the canvas

Take the tightest placement you will actually run on, take its covered regions, and subtract them from the frame. What is left is the rectangle. Everything that has to be read, recognized or understood goes inside it: the proposition, the brand mark, the product at its clearest, the price if there is one. Everything else can live outside it and be beautiful.

This is not the same instruction as "keep your text in the middle." The picture still fills the entire frame. What changes is which part of the picture is load-bearing. A hand entering from the bottom edge is fine. A hand holding the product in the bottom edge is not, because the caption block is about to sit on the product.

04Why is letterboxing an ad a kill?

Because bars are a signal, and the signal is that this ad was made for somewhere else. Black strips at the top and bottom of a vertical placement say the file arrived from a 16:9 timeline and nobody re-made it. Pillarboxing says the same thing louder. Our rule is blunt and it is enforced rather than requested: every frame fills the delivery canvas edge to edge, and the delivered file gets re-measured for flat bands along its edges instead of being trusted.

There is a second cost people miss. Bars do not only look borrowed, they shrink the picture into the exact region the interface is already competing for. You lose height at both ends and then hand what is left to a caption block. The rescue makes the original problem worse.

05Crop the frame, not the asset

There are two ways to make a picture fit a rectangle, and only one of them is free. You can shrink the asset until it fits, which loses resolution and creates bars. Or you can move the frame across a picture that is bigger than the frame, which loses nothing and changes the composition on purpose.

So we composite into a buffer larger than the delivery canvas. Not much larger, and the exact ratio is build configuration rather than doctrine, but the principle transfers to any toolchain: give yourself room to move the frame without resampling it. Reframing for a second placement then costs a crop decision instead of a rebuild. It is the same instinct behind shooting wider than you cut.

06The layout contract: decided in code, before anything is drawn

We learned this the expensive way. A review came back with eight static creatives carrying mechanical layout breakage: text running outside the background plates it belonged to, an image chip almost entirely hidden behind other elements, labels too small to read. None of it was a taste problem. All of it was a system that had no idea a text box had a plate it must not leave.

So it stopped being a review note. A text run that would leave its plate makes the engine grow the plate, or the render fails and says why. Every overlap carries a declared budget and the engine measures the real overlap after compositing, not the intended one. That second half matters: we once shipped animation that was real, expensive, and covered by a type layer in the delivered file. At feed size it read as a bug.

A rule is either in the path, or it is a suggestion.

Our own audit of which rules never broke, and which broke repeatedly

plate       every text run belongs to a named plate
fit         the box fits, or the plate grows. no third option
refuse      a run that cannot fit its grown plate fails the render
bleed       zero letterboxed or pillarboxed frames, edge to edge
buffer      composite larger than you deliver, then move the frame
rect        nothing load-bearing outside the smallest guaranteed rectangle
occlusion   every overlap carries a declared budget, measured post-composite
cap height  >= 5% of frame height, measured off the ink, not the point size
contrast    4.5:1 at delivery width, 3:1 at 360 px, worst local ground
proof       re-measure the delivered file, never the design file

The evidence behind that sentence is unflattering. When we classified our own rulebook, 149 of 293 rules existed only as documentation. The rules that never broke were the handful with a refusing gate in the only path to the effect. The ones that broke over and over were the ones written down as prose, and layout was top of that list. The longer argument is turning taste into checks a machine can run.

07How do I check an ad I already have?

On a phone, in the actual app, as a draft. Not in a preview window on a laptop. The preview lies by omission: it shows you the file, and the file is not what anybody watches. Five checks, about four minutes, and you can do all of them without opening an editor. One of them overlaps with the legibility floors a caption has to clear on a phone, which is the same problem measured on the type instead of on the frame.

One honest limit. Clearing all five makes an ad legal, not good. Format compliance buys a small, real efficiency gain and nothing else: the best number anybody has published is CreativeX's finding of about 2% lower CPM per 10% gain in creative quality score, across 822,000 observations. Worth having. Not a campaign. The campaign is still the hook, which is how a short ad is built rather than how it is exported.

Safe zones get treated as guidelines because they are published as guidance, in a tone that sounds like advice. They are the shape of the hole your work has to fit through. The only difference between a constraint and a guideline is whether something refuses when you break it. Ours refuses. Make yours refuse, or a person has to remember, and people do not. The rest of the pass is in the pre-launch QA checklist for generated footage and in how to brief and QA ad creative.