H3 Max lip sync: why 5.1 seconds of audio is priced as 6 seconds

Sume prices MiniMax H3 Max Lip Sync at ceil(duration_seconds) times the per-second rate, so audio of 5.1 seconds is priced as 6 seconds: $0.60 at 768p, against $0.50 for exactly 5 seconds. The rate at 768p is $0.10 per second, which is the fal list price of $0.08 times Sume's 1.25 margin. Sume documents the surface in its models docs; fal lists the base rates on its model page.

Ceil steps at 768p

Each fraction of a second above a whole second adds a full second to the price.

The 5-14.8 window

The API rejects a duration_seconds outside 5-14.8 with invalid_request and never clamps it. Audio under 5 seconds fails at the provider; audio past 14.8 seconds would be clipped. If a line is 5.05 seconds, you pay for 6; if you can regenerate the speech slightly shorter, you save a second. Do not trim the audio to fit a window artificially: cut lines at sentence ends instead.

Check what you will pay

The submit response includes usage.billable_amount_usd_micros, and the admission preview endpoint accepts the public model id minimax/h3-max/lip-sync, so you can read the number before you commit a batch.

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-ceil-001" \
  -d '{
    "image_url": "https://example.com/host.png",
    "audio_url": "https://media.sume.com/audio/line.mp3",
    "duration_seconds": 6,
    "resolution": "768p"
  }'

curl https://api.sume.com/v1/jobs/JOB_ID/status -H "Authorization: Bearer $SUME_API_KEY"