300 captioned clips on Sume: $60, and a restyle pass costs $60 more

Three hundred clips of up to 60 seconds cost 300 x $0.20 = $60.00 to caption on Sume, because each standalone caption job reserves and captures a fixed $0.20 under the current estimate. A restyle is still a render and costs $0.20 again, so burning the same 300 clips in a second style is another $60.00, for $120.00 in all.

The price is from the Sume video captions docs and the request contract in the repository (read 2026-10-09). The docs tell you to confirm the live rate in GET /v1/catalog and the OpenAPI document, so treat the $0.20 as the figure at the read date.

What the $0.20 covers

The fixed estimate applies to videos of at most 60 seconds. The standalone job takes a public HTTPS video_url (or a source_caption_id to restyle), a style, an optional font, language, and one wording source. It returns a captioned video.

Inline captions on Avatar Video are a different add-on in that estimate and do not create a video_caption resource.

Saving on rework

The restyle path does not run speech-to-text again. It reuses the source video and the word timings of the first caption, so you pay for the render and not for a second transcript. The docs say that sending words with a restyle is only for correcting wording.

Since the price is per job, the cheap rework is to decide the style on a small sample. Caption 10 clips, pick the style, and then run the other 290 once: 300 x $0.20 = $60.00 with no restyle bill. Compare that to the plan where all 300 get a second pass.

Failures and next steps

A speech-to-captions job on a silent clip fails with caption_no_speech. Send cues or segments with text, start, end instead; the contract caps those at 200 items and 60 seconds. If you send script_text and Sume cannot align it to the speech, the job fails with script_alignment_mismatch or script_alignment_failed; remove script_text or simplify it.

Poll GET /v1/jobs/:id/status for each job, or use a webhook for a batch, and send an Idempotency-Key on each submit so a retry does not bill twice.

Choosing a wording source

Pick one of four: speech-to-text by default, script_text to lock your wording over the speech timings, words for your own word timings, or cues for phrase-level overlay text. The price is the same for all four. The choice is about accuracy and failure modes, so for a 300-clip run where the wording is already known, script_text or words avoids a mishearing, but a script_text that cannot align fails the job.

Run a sample of 10 through your chosen path before you submit the other 290.