MiniMax H3 Max lip sync API: a still plus audio, 5 to 14.8 seconds

To make a still image speak with MiniMax H3 Max on Sume, send POST /v1/minimax/h3-max/lip-sync with an image, your audio and its duration_seconds. The audio must be 5 to 14.8 seconds, and the output length follows the audio. It is a separate endpoint from POST /v1/videos, and it does not take a model field.

Facts are from the Models overview docs, the request schema and pricing code, read 2026-09-29.

What does the request take?

The same still-plus-audio body as Sume's other talking-still endpoint. Exactly one visual source is required, and the audio must be hosted by Sume.

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 '{
    "image_url": "https://example.com/presenter.png",
    "audio_url": "https://media.sume.com/your-workspace-audio.wav",
    "duration_seconds": 9,
    "resolution": "768p"
  }'

What are the rules?

Why refuse audio outside 5 to 14.8 seconds?

The provider rejects audio shorter than 5 seconds and silently clips audio past 14.8, which would drop speech. Sume refuses both edges at admission instead. Split longer speech into pieces and join the clips afterward.

When should I use video generation instead?

When you want the model to invent the scene and the sound. With minimax-h3 or minimax-h3-max on POST /v1/videos you can also send an audio reference to steer the voice; see voice reference with audio clips. Lip sync fits when the words are already recorded and must match exactly.