Creatify Boreal-H3 runs 5 to 15 s; Sume H3 Max lip sync stops at 14.8

Creatify's API changelog lists a Boreal-H3 model for clips of 5 to 15 seconds, while Sume's H3 Max lip-sync route accepts 5 to 14.8 seconds and rejects anything outside, including exactly 15. If you are porting a 15-second job, trim 0.2 seconds from the audio before you submit, or the request fails. The route never rounds a number down to fit.

Side by side

The Creatify rows are what its changelog says; I did not test its API. The Sume rows are from the repo's lip-sync doc.

Trim a 15 second line

Sume also bills on whole seconds, so a 14.8 second line bills as 15 seconds. The trim costs you nothing in price, only 0.2 seconds of audio. At 768p the line costs 15 x $0.10 = $1.50.

• Measure the audio length first.
• If it is above 14.8, shorten the text by a few words and render the voice again; do not cut the finished file.
• Set duration_seconds to the new length, rounded to one decimal.
• Submit to POST /v1/minimax/h3-max/lip-sync with the same still.
• If you cannot shorten it, split the line into two segments of at least 5 seconds.

A caution on the comparison

I do not know from the changelog whether Boreal-H3 is a lip-sync model, a general clip model or something else, so I do not compare quality. What the two share is a short window with a hard upper limit, which is the thing that breaks a migration. Check the model's page for what it does before you plan around it.

Cost at the edge of the window

Here is what the limit means in dollars on Sume. A 14.8 second line and a 14.0 second line both bill 15 and 14 seconds, so the difference is one billed second: 10 cents at 768p and 20 cents at 1080p. Tightening a script to fit therefore saves little money; the real value of trimming is that the request is accepted at all. A batch of fifty 15.0 second lines, all rejected, costs nothing but returns no video, so test the first line before you queue the rest. The same preflight is also a good place to check the still: its width over its height has to sit between 0.4 and 2.5.

What Sume does not do

Sume does not extend the window, and it does not clip the audio for you. For lines longer than 14.8 seconds the packet guidance is to use Fabric, which accepts up to 300 seconds, or to split at a sentence end.