MiniMax H3 Max lip sync audio_url: Sume media host only

The audio_url for POST /v1/minimax/h3-max/lip-sync must be a public HTTPS URL on Sume's media host, between 5 and 14.8 seconds, and at most 10 MB. Sume's OpenAPI says non-Sume hosts are rejected. The usual source is a Sume text-to-speech segment, which already lives there.

What are the audio rules?

The schema states them in one description: public HTTPS, on the Sume media host, 5 to 14.8 seconds, 10 MB maximum. duration_seconds is a second required field in the same window, used to reserve credits at admit.

Sume refuses out-of-window requests instead of clamping them. The description gives the reason: the provider rejects shorter audio and silently clips longer, so a refusal is safer than a clip that stops mid-sentence.

What if my audio is somewhere else?

A link from another storage bucket will not pass. The schema names only the Sume media host as accepted and does not describe an upload step for this route, so generate the narration with Sume TTS, which the description calls typical, or check Sume's models overview for the current audio path.

The models overview says MiniMax H3 Max Lip Sync takes the same still-and-audio body as Fabric.

What about audio longer than 14.8 seconds?

Split it first and make one clip per segment. Lip sync longer than 15 seconds walks through the split, and TTS sentence segments covers cutting at sentence boundaries so no clip ends mid-word.

Keep each segment a little above 5 seconds if the script allows; shorter audio is refused.