Highlight a phrase and the clip covers those exact words

Before you touch anything, the machine has read your transcript. It hunts the words doing concrete work. A product, a number, a place, a result, a before and an after.

Those get covered. Abstractions mostly do not. A stock shot over the word therefore is noise with a licence fee.

It then searches for footage matching those words. Each clip is cut to the run of words it sits on. If a clip is shorter than the phrase, nothing is stretched or slowed. It ends, and you are back on camera.

Your own footage gets first pick. Your product, your premises, your faces, used wherever the words fit. Licensed public footage fills the rest.

When the machine picks wrong, and it will, swapping is one action and costs nothing. Replace it, or search millions of free clips for the shot you had in mind.

Watch anybody place b-roll by hand and the ratio is the same. Two minutes to search. Twenty to place.

The twenty minutes are not craft. They are arithmetic done by a human with a mouse. They exist because the timeline thinks in seconds while you think in words.

Nobody has ever wanted a shot to begin at four seconds and thirty-one hundredths. They wanted it to begin on the word everything. So you translate an intention into a number, drag until it is close, then fix the translation error eight times an ad.

Multiply by eight shots, then by every variant you meant to test. That is where the month went. It was never a shortage of ideas.

Marking words removes the translation step. There is no number to get right, so there is no number to get wrong on the second version either.

Coverage is capped by the pace you choose. 40 percent at chill, 45 at normal, 52 at fast. Even at the busiest setting you are on camera for close to half the ad.

People argue with that number before they test it, so here is the reasoning.

An ad that is 80 percent stock footage stops reading as a person talking to you. It starts reading as an ad. That is the moment the thumb moves. The face is the trust, and the footage shows what the face is talking about.

The cap also protects the through-line. Cut away too often and the viewer never settles back into who is speaking, so the argument never accumulates.

Once the ad is at its cap, a further highlight replaces a weaker clip rather than adding one. A ninety second ad at normal pace carries about forty seconds of footage, which is roughly eight shots.

It finds and places real clips. Yours first, then licensed public ones. If your line is about a product that does not exist yet, no prompt invents a shot of it. Upload your own frames instead.

It reads what you said, so what you said decides the results. A line like it changed everything gives the search nothing to grip. Name what changed and the clips improve at once.

Concrete words get good clips. Vague words get generic ones. That is a script problem wearing a footage costume, and it is a thirty second fix on the page.

Everything hangs on a talking-head take. The transcript is what the shots are placed against, so the read is the spine of the whole ad.

And coverage stops at the cap for the pace you picked. Half the ad is the ceiling by design, because the other half is the reason anybody believes it.

Transcription, the directing pass and the footage search are one batch at 100 credits. That covers the searching, the choosing and the cutting-to-words for every clip in the ad.

Swapping a clip afterwards costs nothing. Export is 20 credits per minute of finished video, so a thirty second ad is about 10 credits to download.

The trial is 7 days and 300 credits with no card. Basic is $39.99 a month for 2,500 credits. Premium is $79.99 for 5,000. A $15 top-up adds 1,000 credits mid-month.

Compare that with the resource you were spending, which was never money. Eight shots at twenty minutes each is most of a working afternoon per ad.

An afternoon per ad is how an account ends up testing six creatives a month against a 5 to 8 percent hit rate. The credits were never the constraint. The afternoons were.