2026-10-07

Inspect a 3-minute Short with a transcript: set duration_seconds 180

video-inspect reserves one minute of speech-to-text unless you send duration_seconds (max 600). Why a 180 hint matters, the $0.01 per minute rate, and the 400s.

5 min readSume

All posts

When you ask Sume video inspect for a transcript with transcribe: true, the call reserves one audio minute unless you tell it the length. Send duration_seconds with the clip's length (for a three-minute Short, 180), up to a maximum hint of 600 seconds. The speech-to-text rate is $0.01 per audio minute, so a 180 s clip is about $0.03 of transcript at that rate.

What the docs say, exactly

Three statements carry the weight. The public STT rate is $0.01 per audio minute. Without duration_seconds, Sume reserves 1 minute. The maximum hint is 600 seconds. The docs do not spell out the formula that turns the hint into a reserve, so confirm the number you expect with GET /v1/catalog and by reading the billed amount on the first job. What they do say about inspect as a whole is that it reserves its Modal compute ceiling, captures its own container seconds, and never captures more than the hold.

video-inspect transcript fields and limits, from the Sume docs (read 2026-10-07)
Field	Rule
`transcribe`	true to run Sume STT 1.0
`duration_seconds`	hint for the reservation, max 600 s, only with `transcribe: true`
`language_code`	STT hint such as `en`; auto-detect if omitted
`segmentation.mode`	`sentence` returns gapless sentence `segments[]`
`silence_split_seconds`	0.2 to 3
Rate	$0.01 per audio minute
Source length	up to 1800 s

Why the hint matters for Shorts

A Short under one minute fits the default reservation. A 180-second one does not, because the transcript is three minutes of audio. If the reservation is smaller than the work, you would rather learn that on the first call than on a batch of a hundred. For clips over 600 seconds the hint tops out, so a very long recording is better handled by detaching the audio and using the STT route directly.

A request

Sentence segmentation returns lines you can use as caption cards. If the clip has no audio, the call fails with inspect_source_has_no_audio, so check probe.has_audio first with a frames: false inspect.

{
  "video_url": "https://media.sume.com/artifacts/artf_demo/short.mp4",
  "frames": false,
  "transcribe": true,
  "duration_seconds": 180,
  "language_code": "en",
  "segmentation": { "mode": "sentence" }
}

Common 400s

Sending language_code, segmentation or duration_seconds without transcribe: true returns video_inspect_transcribe_required. The default mode is sync with a 30-second wait; if it does not finish, you get a 202 and poll the job.

Sources

Sume docs: video-inspect