Captioning 200 short videos: $40 at $0.20 a job, up to 60 seconds

Captioning 200 short videos costs $40.00 on Sume, because a caption job is a fixed $0.20 for a video up to 60 seconds long (read 2026-10-06). The price is per job, not per second, so a 6-second clip and a 59-second clip cost the same. That makes the cost of captions a function of how many clips you caption, and a re-caption after an edit is a new job at the same price.

The question for a library is what to caption first. Captions on a 6-second clip cost as much as captions on a 59-second one, so the shortest clips have the highest cost per second and the least to read. Caption the long ones first if you have a budget to ration, and the short ones when they are going to a platform that shows text in the feed.

Cost per second changes, cost per job does not

At a flat $0.20 a job, a 10-second clip is $0.02 a second, a 30-second clip is about $0.0067 a second and a 60-second clip is about $0.0033 a second. If you can, caption a finished 55-second cut rather than five 11-second parts, since five parts cost $1.00 and one cut costs $0.20. The same logic argues for captioning once, after the cut is final, because every trim or reorder invalidates the caption timing.

That order of work is the largest saving in the whole workflow. Trim and compose steps are billed at $0.02 each, so do the edits first, check the picture, and caption last.

• Caption the final cut, not the parts.
• Check the clip is 60 seconds or shorter before you submit, or expect to split it.
• Do not re-caption for a typo: fix the text if your pipeline allows, and budget one re-caption in 4 as a planning figure, not a measurement.
• Failed jobs are refunded.

Price a library

The function prices a list of clip durations and flags the ones over 60 seconds, which need a different plan.

def caption_plan(durations: list, recaption: float = 0.0) -> dict:
    ok = [d for d in durations if d <= 60]
    long = [d for d in durations if d > 60]
    jobs = len(ok) + round(len(ok) * recaption)
    return {"jobs": jobs, "cost": round(jobs * 0.20, 2),
            "needs_split": len(long)}


clips = [8, 15, 30, 45, 59, 75] * 33 + [20, 20]
print(caption_plan(clips))
print(caption_plan(clips, 0.25))

A library audit before you spend

Before captioning a backlog, sort it. Videos that will be watched with sound on, such as a product walkthrough on your own site, gain little from burned-in captions, so leave them out. Videos for a feed where people scroll muted need captions, and the ones with the most views deserve them first. A simple order is to rank the library by expected views, caption the top 50 at $10.00, look at what changed, and decide on the rest.

Keep a ledger of which clip has captions, which version of the cut they match and which job produced them. When someone trims a clip, the ledger tells you the old captions are stale and that a re-caption of $0.20 is needed. Without it, stale captions ship, and that is a worse cost than the twenty cents.

The same thinking applies to batch size. Submit 24 captions at a time on a Pro plan, wait for them to finish, check two or three by eye and then send the next wave. A systematic fault, such as captions sitting too low for a platform's interface, then costs you one wave of $4.80 and not the whole $40.00.

Checking before a big run

Run a pilot of five clips, spend $1.00, and read the output against the audio. Check the line breaks and the reading speed on the phone screen, not on your editor. Then submit the library in waves that fit your plan's accepted-job capacity, which is 6 for Free, 24 for Pro, 48 for Startup and 120 for Scale. A queue_full response means wait, not retry harder.

The details are in the captions docs, the pricing page and the billing docs.