A 4-second spoken line: H3 Max Lip Sync refuses it, Fabric $0.75

A 4-second spoken line cannot go to MiniMax H3 Max Lip Sync on Sume, because the route accepts audio of 5 to 14.8 seconds only. Send it to VEED Fabric 1.0 instead: 4 x $0.1875 = $0.75 at 720p, or 4 x $0.10 = $0.40 at 480p (as of 2026-10-09).

This matters for short interjections: a greeting, a product name, a one-line sign-off. They are real clips that people generate, and they are exactly the length the cheaper lip-sync route will not take.

The rule

Sume's models page lists the H3 Max Lip Sync body as the same still plus audio body that Fabric uses, with audio of 5 to 14.8 seconds. The API validates the declared duration and rejects values outside that range. It does not stretch or pad the audio for you, because silently changing the audio would change what the speaker says. A request with a 4-second duration fails with an invalid_request error.

Fabric has no such floor. Sume's Fabric route accepts durations from 1 second upward, so a 4-second line is a normal Fabric job.

Do not pad the line to reach 5 seconds

The tempting fix is to append a second of silence so that the file reaches 5 seconds and the cheaper route accepts it. That changes the audio file you recorded, and the clip then ends with a speaker who stares at the camera in silence for a second. If you need that pause anyway, add it deliberately; if you do not, pay the Fabric price. The cost difference is small: a padded 5-second H3 Max clip at 768p would be 5 x $0.10 = $0.50, against $0.75 for Fabric at 720p, a gap of $0.25.

A greeting library

A set of 12 greetings of 3 to 4 seconds each is a good use of Fabric. At 4 seconds each, the 12 clips are 12 x 0.75 = $9.00 at 720p, or 12 x 0.40 = $4.80 at 480p. Because there is no 5-second floor, you can use the same still and the same handle for all of them and render them in parallel, subject to your workspace queue.

A line that sometimes comes out at 4.6 seconds and sometimes at 5.2 is awkward for H3 Max. Whether it is accepted depends on the measured audio, not on what you intended. Decide per file, after measuring, which route gets it.

Fabric's price in context

Fabric's vendor page lists $0.08 per second at 480p and $0.15 at 720p (read 2026-10-09). Sume bills provider list times 1.25, which gives $0.10 and $0.1875. For 100 such greetings, 100 x $0.75 = $75.00 at 720p and 100 x $0.40 = $40.00 at 480p. Sume's models page also lists a longer one-off ceiling for Fabric clips, so a script that sometimes produces 4-second lines and sometimes 20-second lines can use Fabric for all of them and avoid a router.

Remember that the 14.8-second upper limit does not apply to Fabric in the same way, so mixed-length scripts are simpler on one route.

• Under 5 seconds: Fabric.
• 5 to 14.8 seconds: either route; H3 Max at 768p is $0.10 per second.
• Over 14.8 seconds: Fabric, or split the script at sentence boundaries before you record.