Review a draft: Sume caption job $0.20, or transcript $0.01/min

Short answer

Both routes read the speech in a draft clip, but they bill differently and return different things. A standalone caption job reserves and captures $0.20 for videos of up to 60 seconds under the current fixed estimate (Video captions). A video-inspect call with transcribe: true adds a per-minute rate of $0.01 per audio minute on top of the compute reserve for the inspect itself (Video inspect).

What each one costs on a draft

For a 30-second draft, the two numbers are not directly comparable, because the inspect call also bills its Modal compute and the caption job is a flat fee. The table shows the parts the docs state.

When each wins

Use the caption job when the deliverable is a captioned video. Use the inspect transcript when you want the words, the stills and the probe in one call, for example to check that a draft says the line you scripted before spending on the final render. A silent clip gives inspect_source_has_no_audio, so check probe.has_audio first with a frames: false inspect.

Both prices are fixed estimates in the docs, and the docs tell you to confirm the live price in GET /v1/catalog.

Cost guard

• Send duration_seconds on the inspect call; without it Sume reserves a full minute.
• Caption only the draft you intend to ship; a captioned draft is not reusable as the final.
• Failures release the reserve, so a rejected request does not stay charged.

A worked review pass

A weekly series has twelve 45-second drafts. Captioning every draft with the standalone job costs 12 x $0.20 = $2.40. A transcript through inspect with a 1-minute duration hint adds $0.01 per clip for speech-to-text, $0.12 in total, plus each inspect's compute, which the docs bill at container seconds x Modal list x 1.25 plus the platform fee and cap at the hold. Compute is small for a short probe but is not zero.

The cheaper habit is to transcribe the drafts, review the words, fix the script, and caption only the final cut.

Common mistakes

Sending language_code, segmentation or duration_seconds without transcribe: true returns 400 video_inspect_transcribe_required. A caption job for a clip over 60 seconds is outside the flat price, so split first. And remember the docs' caution that these are fixed estimates; read the live price in GET /v1/catalog before you build a budget.

The flat price is attractive for short clips and awkward for very short ones, and the per-minute transcript is the opposite. A 10-second draft costs $0.20 to caption but a cent or so to transcribe, so for a quick wording review the transcript wins by a wide margin. For a final 55-second cut where you want styled captions burned in, the flat $0.20 is the better deal.

Keep one rule in mind: decide what you need from the pass. If you only need to read the words, transcribe. If you need caption output you can ship, caption once, on the version you approved.