Doing the arithmetic before a call is spent
Speech runs at a measurable rate and generated shots have published ceilings. Between the two there is a short sum that decides how a line has to be written, and doing it at the writing stage is the only cheap moment. As of 2026-09-12.
| Figure | Where it comes from | What it decides |
|---|---|---|
| A pause at each end | How a shot is cut, not a model | How much of the shot is speech |
| A shot ceiling | The vendor documentation | How long one generation runs |
| A speaking rate | Published measurement | How many words fit that length |
Inclusion rule. Figures required to size a spoken line against a generated shot, and where each one is found. Order. Alphabetical by figure.
1Three words a second, minus the pauses
At roughly three words a second, a six-second shot holds about eighteen words including the beat before the line and the beat after it. Dialogue written without that constraint reliably runs longer.
The consequence is either a rushed read or a line that does not fit the shot it was written for. Both are rewrites, and at generation time a rewrite also costs a call.
2The pauses are part of the count
Measured speaking rates that include silences run well below rates that exclude them. A shot list budgeting words at the faster figure produces lines with no room to breathe, which reads as hurried even when every word technically fits.
Leaving half a second at each end is a useful default and it is a cutting decision rather than a model property. It survives a change of platform, which most of this arithmetic does not.
3A fixed menu of lengths is easier to write to
Where a vendor publishes four, six or eight seconds, a writer can work in those units and everybody involved sees the shape of the thing being made. A wide range postpones the decision and moves the cutting to a worse moment.
Where a total is reached by extending a generation, the figure is a budget for an assembly rather than a take. Sizing a scene to it produces a line that spans a join, which is where a voice moves.
4Where the published detail is logged
What is described above is how the techniques work in general. Which models document which of them, in whose words, is kept in the speech controls table with a date on every field.
- Dialogue timing — the published figures behind the sum
- What sets the length — a menu, a range, or a take
A mechanism note rather than a documented field. Nothing here is attributed to a model, and nothing here is a claim about one. The sourced material is on the speech controls table. Related: A cast list, Writing the lines.