An hour of calm costs the same as a breath.

Wellness audio is long by design — a sleep story runs half an hour, a course of meditations runs for weeks. Per-character pricing turns the soothing length that is the whole point of the product into the biggest line on the invoice.

2,000

characters a request — an hour is a loop over them

The arithmetic of soothing length

A single request carries up to 2,000 characters and streams gap-free, so a guided session is a loop and a full sleep track is a longer loop. On an unmetered seat the loop costs the same, whatever its length, and the running time that defines the product stops being a meter reading.

The practical effect is on tone. Calm is a thing you tune by ear, and re-rendering a track that came out a shade too bright is free here, so a track gets re-voiced until it lands rather than shipped because the budget is spent.

The controls a calm read needs

temperature narrows or widens the melodic range — measured 10.1 to 18.0 semitones across its span — so a sleep track can sit flat and even rather than perform.
cfg_weight loosens the pacing, so a breath cue gets the slow, unhurried delivery it needs instead of being hurried through.
Break tags place the silence between cues exactly where the script wants the listener to settle.
One cloned voice holds across 23 languages, so a meditation keeps the same guide in every localised edition.

Notes — an engineer's checklist

01How do I keep the same guide across a long series of tracks?

The voice is a request parameter, so every session is conditioned on the same reference clip and there is no drift between tracks or sessions.

02Can I re-voice an existing track for a calmer tone?

Yes. Lower the temperature to flatten the melodic range and ease cfg_weight for slower pacing, then re-render. Re-rendering costs nothing on the seat, so tone is something you iterate on rather than settle for.