A 20,000-character voiceover is about 22 minutes: render bill $2.30

One Sume TTS job takes up to 20,000 characters and costs 20 x $0.0475 = $0.95. At an assumed 15 characters per second of speech, that is 1,333 seconds, about 22.2 minutes. Timeline render reserves ceil(audio.duration_seconds / 60) minutes at $0.10 each, so that spine reserves 23 minutes: $2.30. The whole voiced video is about $3.25 before any captions.

The chain of numbers

The 15 characters per second is an assumption, not a Sume figure: it is a plausible pace for English narration, and your script, language and voice will move it. Measure one finished file and replace it. Everything else comes from the docs and catalog: TTS maximum 20,000 characters and $0.0475 per 1,000, Timeline audio.duration_seconds from 1 to 1,800 and $0.10 per ceil(output minute).

Why ceil matters

The render reserve rounds the spine up to the next whole minute, so a spine of 1,321 seconds (22.02 minutes) costs the same as one of 1,380 seconds (23 minutes). Trim the last seconds of silence and you can drop a minute, a 10-cent saving. Run the unbilled POST /v1/timeline-1.0/plan call first; it returns billable_minutes and estimated_cost_usd_micros without creating a job.

Staying inside the caps

Three caps decide whether the plan is valid.

• The spine can be at most 1,800 seconds (30 minutes). At 15 characters per second that is 27,000 characters, so a longer script needs two TTS jobs and a timeline audio concat at $0.01.
• The video list can have 1 to 200 slots, so a 22-minute video of 5-second shots needs about 267 slots and does not fit; use longer slots or stills held for 10 seconds.
• Every URL must already be this workspace's media.sume.com file. Import first.

What 20,000 characters is not

Do not read 22 minutes as a promise. The 20,000-character cap is a request limit, not a length. A script of short words, many numbers or pauses written as punctuation runs at a different rate, and the voice you choose changes the pace again. Treat the spine length as the measured length of the file that comes back, and put that number into audio.duration_seconds, which can be 1 to 1,800.

Also note what is not in the $3.25: captions are a separate $0.20 job for each clip up to 60 seconds, so a 22-minute video needs a plan for captions rather than one call. Music is a separate $0.125 generation per track.

Fast check

Before you submit, call plan with the real spine length, check billable_minutes, and compare it with your own estimate. If they differ by a minute, your characters-per-second guess was off, and the guess is the thing to fix.