How a short ad is built: shots, sound and type

01How long should a shot be in a 15-second ad?

Between roughly 0.2 and 2.5 seconds, and the median lands near 0.8. Across 48 short ads we measured cut by cut, the median mean shot length is 0.83 seconds; product-led films cluster tighter, at 0.13 to 0.35. The one film in that set that was rejected outright did not make its first cut until 1.08 seconds.

That last number is the one to sit with. By the time that film cut once, the fast ads beside it in the same feed had cut somewhere between five and thirty times. It was not slow because it was calm. It was slow because nobody had decided what any individual shot was for, so every shot ran until the next one was ready.

We used to run a hard ceiling of two to three seconds on any generated shot. It came out of one bad clip and it did not survive contact with a good one. A generated shot that genuinely holds, because it has an arc or its own scene cuts inside the take, runs at whatever length the direction calls for. Duration in a generator is a pace instruction, not a length, which is a separate trap and a longer story.

Pace is the mechanism. Fast cuts hold attention and deny the eye time to find flaws.

Our own house note on why generated footage gets cut fast

02The canvas is a decision, and it is the cheapest one you make

Every film we ship is 1080 by 1920 at 24 frames a second, full bleed, with zero letterboxed or pillarboxed frames anywhere in it. At 24fps one frame is 41.7 milliseconds and fifteen frames is exactly 96 beats per minute, which is the cleanest integer grid available to you. Put the cuts on whole frames against that grid and the sound lands for free.

canvas     1080 x 1920, 24 fps
codec      h264, crf 18, bt709
bleed      zero letterboxed or pillarboxed frames
grid       cuts on whole frames; 15 frames = 96 BPM exactly
sound      present at frame 0, never a fade-in
end        final 2s at >= 0.60 of the peak 2s RMS
type size  cap height >= 5% of frame height, measured off the ink
type dwell max(1.2s, 0.35s per word), at full opacity, fades excluded
contrast   4.5:1 at delivery width, 3:1 at 360 px, worst local ground

None of that is styling. It is the set of things that must be true before anyone can have an opinion about the film. Sound at frame zero is the most corroborated finding we hold: across 133 films in three separate studies, we found no exception. The film we once scored 5 out of 100 took 0.178 seconds to reach first sound and was still ramping at 0.9 seconds. It spent the entire scroll decision in silence.

03How many cuts fit into 15 seconds?

About twelve to fifteen. Our own reference set runs 14 to 18 cuts across 17 to 18 seconds, so a 15-second film in the same idiom lands around twelve to fifteen. Use it as a sanity check and not as a brief, because we published that range as a target once and then withdrew it: twenty films built toward a cut count came back as one film made twenty times.

Later we retired eight finished ads in a single afternoon over the same disease, and the lint that measured editing mechanics went in the bin that day. What replaced it measures perception: no two pieces in a set may match on more than two of the six axes a viewer actually reads. What makes an ad resist fatigue is the long version.

What survived the retirement is not a count. It is four pacing habits: a shot ceiling, a motif that runs no more than three shots before the film has to change its mind, a beat of color every few seconds, and no two adjacent shots sharing a scale class. That last one does most of the work. Macro, then macro, then macro reads as one long shot no matter how many times you cut it.

04One film, second by second

Here is the whole thing in one picture: an 18-second film with picture, sound and type on their own rows, so you can see the moments where all three are doing different jobs at once. The picture row groups the beats rather than showing all sixteen cuts, because the argument is about what each window is for.

The second thing to notice is at 8.6 seconds. Any film over about ten seconds needs a second hook, and it needs to be declared in advance with the frame it lands on. In several films we measured, the re-hook carries more production value than the opening. It has to work by a different mechanism than the first hook, or it reads as the film repeating itself.

05Where does the brand go?

Between 20 and 30 percent of runtime, which on an 18-second film is 3.6 to 5.4 seconds in, after the hook has resolved and never inside it. The end card is a repeat, not an introduction. That window is ours, and it is a synthesis of two measurements rather than a measurement itself, which is worth saying plainly.

Those two bars look like a contradiction and they are the same rule at two different runtimes. The 1.3 seconds is one film: a single luxury reference we timed, in which a wordmark everybody already knows lands early and reads as a recognition event, because your eye hits it and something happens. An unknown brand's wordmark in the same position is a piece of text nobody can read yet, which is homework rather than a stimulus.

One real exception. A quiet single-claim film, the kind that makes one argument and stops, is often correct to hold the brand to the payoff and put nothing before it. What is never correct is arriving at either answer by accident. The direction lock names the beat the brand appears on, in writing, before anything renders.

06The build order, and the three places changing your mind is free

This order exists because of a week that scored 2 out of 100, when every correction was landing on a finished render instead of on paper. Why we get paper approved before pixels tells that story properly.

All of this makes a film that is allowed to be judged, and nothing more than that. The number that keeps me honest is 43 percent: fewer than half of our films that clear every automated check survive the first human watching them. The checks are the floor. Nobody has ever bought the floor.