use case
verified July 2026
Cut fifty reads, keep the one that converts.
A spot is not written once. It is written fifty ways — the hard sell and the soft one, the fifteen-second cut and the thirty, the version for each market — and the read that converts is only known after they have all aired. A meter charges for every version of a decision you have not made yet.
01
Fifty reads, one that airs
The creative process for a spot is elimination. You generate the alternatives — the tones, the lengths, the calls to action — and you cut everything but the one that tests best. A meter prices the forty-nine you throw away exactly like the one you keep.
On an unmetered seat the arithmetic inverts. Generating every variant costs the same as generating one, so which read converts becomes a test you can actually run across markets rather than a bet placed before the budget runs out.
02
The controls a spot needs
- Instant voice cloning holds a booked-talent read as a request parameter, so a pickup line or a new call to action lands in the same voice weeks after the session.
- temperature sets melodic range, so one script reads as a hard sell or an aside without booking a second voice.
- Twenty-three languages from one seat, so a spot localises into each market instead of commissioning a session per territory.
- Break tags place the beat before the tagline exactly where the edit wants it, and a corrected line re-renders at no cost.
03
Notes — an engineer's checklist
01Can we localise the same spot into other markets?
Yes — the script is a request in any of twenty-three languages and the voice is a parameter, so a market gets the same read localised rather than a fresh casting brief and a new session invoice.
02Can we reuse a voice actor's read for later pickups?
Yes — five to ten seconds of the talent is enough, and the reference rides in the request. Cloning requires a consenting speaker, and every clip we synthesise carries an inaudible provenance watermark applied at generation.
03Is a synthetic read good enough to air?
That is your judgement to make on your own script, which is why the console on the home page runs the production API on whatever you type. We publish the measurements and the audio rather than an adjective.
See also — related sheets
capability
0
Voice cloning from ten seconds, in the request
Zero-shot voice cloning: a ten-second reference rides inside each request, no training job, and the identity holds across 23 languages.
capability
7.9
Delivery is a request parameter
The prosody dial moves measured pitch range from 10.1 to 18.0 semitones on one voice, one sentence — continuous per-request control on the production API.
use case
1
Ship the voice in every market, not the top three
Localised product audio on a flat seat: one cloned identity across 23 languages, with the character count no longer deciding which markets get voice.
use case
107 ms
A show that can be re-cut after it is recorded
Podcast audio on an unmetered seat: intros, ad reads, corrections and localised editions re-render as often as the show needs, at no cost per character.
A key and one seat to build this on — the same production API this page measures.
Request access for this use case