9-second voiceover with MiniMax H3 Max lip sync: $0.90 or $1.80

A 9-second voiceover through MiniMax H3 Max lip sync on Sume costs $0.90 at 768p and $1.80 at 1080p. The route bills ceil(duration_seconds) times the provider list per second, times 1.25: 9 times $0.08 times 1.25 at 768p, and 9 times $0.16 times 1.25 at 1080p. The audio must be 5 to 14.8 seconds long, so 9 seconds is inside the window.

MiniMax released H3 on July 31, 2026, with native stereo audio and a 768p price tier (read 2026-10-07). This route is different from H3 Max in the Video Router: it takes a still and audio, and you do not write a prompt.

The price for a 9-second line

Sume's rate card for this route is the fal lip-sync list of $0.05 at 480p, $0.08 at 768p, and $0.16 at 1080p per second (fal page, 2026-09-22), times 1.25. There is no 2K option. The default resolution is 768p.

What the route needs

The audio sets the output length. fal rejects audio under 5 seconds and clips anything after 14.8 seconds, which is why Sume refuses to guess.

• A visual source: one of image_url (aspect ratio 0.4 to 2.5) or a ready avatar by avatar_id or avatar_handle.
• An audio_url hosted on Sume, no larger than 10 MB, 5 to 14.8 seconds long.
• A measured duration_seconds from 5 to 14.8. Sume refuses values outside that range and never clamps them.
• resolution: 480p, 768p, or 1080p. speed_tier is accepted and ignored.

A request

Generate the voiceover first with the TTS route, so the file is on Sume's media host, then pass its URL and measured length.

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: lipsync-demo-001" \
  -d '{"image_url": "https://example.com/presenter.png", "audio_url": "https://media.sume.com/artifacts/artf_demo/line.mp3", "duration_seconds": 9, "resolution": "768p"}'

768p or 1080p

1080p doubles the price: $0.90 more on a 9-second line. Use 768p for drafts, social vertical clips, and anything viewed on a phone, then spend the extra on 1080p only for the take you ship to a larger screen. The reservation is captured on completion and refunded on failure, so a bad still costs a retry, not a charge.

If your line is longer than 14.8 seconds, split it into two lines before you generate the audio. The 5 to 14.8 second window explains how to measure the file.