13.4-second voice-over: H3 Max Lip Sync bills 14 s, $1.40 at 768p

A 13.4-second voice-over costs $1.40 on Sume's MiniMax H3 Max Lip Sync at 768p, because the price is ceil(13.4) = 14 seconds times $0.10 (as of 2026-10-09). At 480p it is 14 x $0.0625 = $0.875. The 0.6 seconds you do not use are still billed: 0.6 x $0.10 = $0.06 at 768p.

How the rounding works

Sume's internal API notes for this model say the charge is ceil(duration_seconds) times the per-second rate, reserved at admission, captured at completion and refunded on failure. The duration you send is the reservation basis, and it must fall between 5 and 14.8 seconds. The audio file sets the real clip length. A value outside the window is refused rather than clamped, so a 15-second file is never silently cut.

Trimming 0.4 seconds saves $0.10

The saving from trimming the voice-over to 13.0 seconds is 1 x $0.10 = $0.10 at 768p, or $0.0625 at 480p. That is only worth doing when the trim removes dead air at the start or end, such as a breath before the first word. Do not cut into words: Sume's guidance for lip-sync runs is to leave the recorded audio intact, because the audio sets what the mouth says.

If you generate the line with Sume text to speech instead of recording it, check the length of the result before you submit. TTS is billed at $0.0475 per 1,000 characters, so cutting 10 characters from a script saves under a tenth of a cent (10 x $0.0000475), while cutting a second of audio saves 10 cents. Cutting words is the cheaper lever.

Planning a batch of odd lengths

The rounding matters most when you have many clips. Take twenty lines that each run 13.4 seconds. At 768p that is 20 x 1.40 = $28.00, while the audio total is only 20 x 13.4 = 268 seconds, which at a flat $0.10 would be $26.80. The rounding adds $1.20 over the batch, 4.5 percent. At 480p the same twenty lines are 20 x 0.875 = $17.50.

The cheapest correction is at the script stage. If a line is 13.4 seconds, cut one clause and re-record at 12.8 seconds; that bills 13 seconds, which saves $0.10 per line at 768p. Measure once, correct once, and keep the measured length in your job log so the next estimate is exact.

Checklist before submit

Measure the file with an audio probe and keep one decimal. Round up yourself to get the billing number. Compare it with the 14.8-second ceiling, because 14.8 itself also bills as 15 seconds, which is $1.50 at 768p.

Keep a note of the billable length next to each file name, for example line-07 = 13.4 s = 14 billable.

• Duration 5.0 to 14.8: send it as duration_seconds.
• Duration under 5: use Fabric (see the rate table in the Sume models page).
• Duration over 14.8: split the script before recording, or use Fabric.