Voiceover too long for H3 Max lip sync: the 5 to 14.8 s window

MiniMax H3 Max lip sync on Sume turns one still and one audio file into a talking clip, but the audio must run from 5 to 14.8 seconds and be hosted on Sume, up to 10 MB. Sume refuses duration_seconds outside that window with invalid_request instead of clamping it, because clamping would cut speech without telling you.

What the endpoint takes

Submit to POST /v1/minimax/h3-max/lip-sync, with one visual source: a public HTTPS image_url whose aspect ratio is between 0.4 and 2.5, or a ready avatar_id or avatar_handle. Add an audio_url on media.sume.com and a required duration_seconds. The resolution is 480p, 768p (the default) or 1080p, with no 2K.

This is not the prompt-driven minimax-h3-max of the video catalog. It is the explicit alternative for when the audio fits the provider's window. The docs keep Fabric as the default talking model.

curl -X POST https://api.sume.com/v1/minimax/h3-max/lip-sync \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: h3-lipsync-001" \
  -d '{
    "avatar_handle": "yura",
    "audio_url": "https://media.sume.com/artifacts/artf_demo/line.wav",
    "duration_seconds": 9.2,
    "resolution": "768p"
  }'

Why the window is strict

The provider rejects audio under 5 seconds and silently clips audio after 14.8 seconds. If Sume clamped your reservation, you would get a clip shorter than your line, and part of the speech would be gone. So the API checks duration_seconds itself and refuses anything outside the window.

What to do with a longer script

Cut the script into lines of 14 seconds or less, generate each voice line, and submit one lip-sync clip per line. Join the clips later on a Timeline render. You have to do the cutting yourself, because the API does not trim audio for you.

If you need a single take longer than 15 seconds, use the avatar talking-video route instead, which plans 4 to 60 seconds.