Narration is the length

PageToVid does not record a separate clip of speech per scene and stitch them together. It records the whole narration as one unbroken take, then finds the natural pauses in that take and uses them as the scene boundaries.

That design choice is why the voice never sounds spliced, never clips a word at a cut, and never drifts out of step with the captions. It is also why a scene cannot be given an arbitrary duration: the scenes partition the voice, and the partition has to add up to the take. Give one scene more time and another loses it.

So the honest statement of the rule is: you set the running time by writing more or fewer words.

The delivered pace is around 110 words per minute — slower than conversational speech, because a voice-over competing with motion on screen has to be intelligible, not fast.

• 15 seconds ≈ 28 words
• 30 seconds ≈ 55 words
• 45 seconds ≈ 82 words
• 60 seconds ≈ 110 words
• 90 seconds ≈ 165 words

Add a few seconds for the opening and closing cards, which carry no narration. If you are briefing an assistant, give it the word budget and not just the seconds — "about 55 words" lands far more reliably than "about 30 seconds".

There is a target_seconds parameter, and it is worth being precise about what it does. It is a brief, not a lever: it tells the planner how much room it has, so it writes a script of roughly the right size and picks a sensible number of scenes.

What it cannot do is compress narration that has already been written. If a script comes in at 200 words, the film is about 110 seconds regardless of what the target said. The planner is measured against the target after the fact, and a take that runs badly over the prediction is re-read — but the words themselves are the constraint.

The most common reason a video feels rushed is not that it is short. It is that a scene carries two ideas, so the visual changes while the second idea is still being explained, and the viewer loses both.

A scene should be one claim, one number, one screen, one moment. If a sentence contains "and also", it is usually two scenes.

This interacts with length in a useful way: since the words set the duration, splitting one crowded scene into two does not make the film longer. It makes the same words land on twice as many pictures, which almost always reads better.

A scene with no narration is the exception to all of this. Having no words to partition, it holds for the duration_seconds you give it.

That is what title cards, end frames and a held product shot are for. Use them deliberately — two or three seconds of silence at the end of a film, with the call to action on screen, reads as confidence. Silence in the middle usually reads as a mistake.

Because the captions are cut from the same measured take, they normally sit exactly on the voice. If a film comes back with them out of step — most often after you have edited the narration — you do not need a new take.

rerender_video with captions_only re-times the captions against the film's existing voice track. Nothing is re-recorded, nothing is re-filmed, and it is free once per render.

If a scene's words take longer to say than the scene was held, that appears in the render's warnings as a sync problem. It is worth reading them before you publish: a scene that is 0.4 s short is invisible in the editor and obvious to a viewer.