H3 Max lip sync and audio over 14.8 seconds: split or Fabric
Audio over 14.8 seconds is refused on Sume's H3 Max lip sync because the provider silently clips the rest. Fabric 1.0 takes the long line.
No: a 20 second voiceover does not fit one H3 Max lip sync job on Sume. The API rejects duration_seconds above 14.8 with invalid_request, because the provider silently clips everything after 14.8 seconds and a clamped reservation would render a clip shorter than your audio, with speech lost. Send the long line to Fabric 1.0, or split it at a sentence boundary into takes that each fall in the 5 to 14.8 second window.
Facts from the Sume H3 Max lip sync API notes on main (origin/main b28f3b0, read 2026-10-13) and the Timeline audio docs.
| Option | How | Watch for |
|---|---|---|
| Fabric 1.0 | One avatar_image_to_video job with the whole audio | Audio must be Sume-hosted and 10 MB or less |
| Two H3 Max takes | Generate two TTS takes of 5 to 14.8 s each | Both must be in the window; one lip sync model per run |
| Split an existing file | POST /v1/timeline-1.0/audio with operation: split and ranges | Keep wav; mp3 adds priming padding at each edge |
How do I split an existing voiceover?
Timeline audio split takes a Sume-hosted file and 1 to 20 ranges, and returns one durable media.sume.com file per range with no re-synthesis. The default output.format is wav (pcm_s16le, sample-exact); the docs say to keep wav if the file drives lip sync. Choose the cut at a pause so no word straddles two takes.
curl -X POST https://api.sume.com/v1/timeline-1.0/audio \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: split-voiceover-001" \
-d '{
"operation": "split",
"url": "https://media.sume.com/artifacts/artf_demo/voiceover.wav",
"ranges": [{ "start": 0, "end": 10.2 }, { "start": 10.2 }]
}'What does it cost?
Each split is one flat-credit job with no model inference. Each H3 Max take is billed by ceil(seconds) at the documented per-second list times 1.25. A take of 10.2 seconds bills 11 seconds.
Sources
Related posts
More in Models
- H3 Max lip sync rejects audio under 5 seconds: what to do
Sume's MiniMax H3 Max lip sync takes 5 to 14.8 seconds of audio and returns invalid_request outside that window. Fabric is the fallback for shorter lines.
- Porting a Sume agent loop to Haiku 5.5: drop temperature, top_p, top_k
Anthropic says to omit temperature, top_p and top_k on Haiku 5.5, and some values return a 400. Strip them from your loop before the first Sume tool call.
- HelixBench: what spatial-acoustic consistency tests in AI video
HelixBench scores whether synthesized sound fields track camera movement. What the arXiv page says it measures, and a manual way to spot-check any AI clip.
- HelixWorld license: Apache 2.0 code, weights and paper differ
HelixWorld's code is Apache 2.0, its weights have a separate license on Hugging Face, and the arXiv page lists CC BY-NC-SA 4.0. Read all three.
Written by Sume