A Human Service Charges $2.20 a Minute for Subtitles. The Machine Does the Minute for $0.006, in Three Languages

Why this pays

A client has a ten-minute video and wants captions in its own language plus two more. The same minute of video has three prices, and they are two orders of magnitude apart.

Ten minutes in three languages is under a dollar of machine time. What the client is paying for, at anything up to $2.20 a minute, is the fix pass: cues that fit the screen, start when the speaker starts, and spell the names right in every language. That pass is yours, and it takes the same twenty minutes whether the client pays you a captioner's wage or a service's rate.

The play, in three lines

1. Ask for cues, not a transcript: Include subtitles on in ElevenLabs Speech to Text, or response_format=srt with whisper-1 on OpenAI.
2. Fix the source cues against the video once, because every uncorrected error will appear in three languages; then translate the SRT as a file with the timecodes locked, by the token.
3. Check each translated file against the line-length rule, export three files named by language code, and put the machine bill on the invoice next to your fix pass.

The path above walks every click. What follows is the reasoning behind each step and the numbers you quote from.

Where the money comes from

• A client with a video and a deadline. Three SRT files, or three plus a dubbed track, priced under the $2.20 a minute a human service charges.
• A platform, if you would rather be paid for the typing. Rev pays captioners $0.54 to $1.10 per audio or video minute, weekly via PayPal within a week of approval, after a waitlist; subtitle translators earn its highest per-minute rates, unprinted.
• Not the provider. Transcription is billed per minute, translation per token, dubbing per minute of source media per language, all from the first call.

The pain you are selling the cure for

The client's video is finished and the platform it goes on wants captions, and the audience in two other countries wants them in their own language. Auto-captions spell the product name three ways and hold a cue for two sentences. The client wants files that pass the platform's line rules on the first upload. That is not a transcription problem; it is a cue problem, and it is why step one is a toggle and not a transcript.

Step one: cues, not a transcript

• A transcript has word-level timestamps, speaker labels and no constraints. A subtitle file has only cue start and end times, no speaker names, no overlapping segments, and rules for characters per line, lines on screen and cue duration.
• On ElevenLabs the two are separate objects in the editor; a change to one does not touch the other. Ask for subtitles at the start.

Path A: in the browser with ElevenLabs Speech to Text

• Transcribe files, upload the video itself, up to 3 GB. Add names, places and product terms as keyterms. Switch on Include subtitles. Scribe v2 is the default model.
• Open the result and switch to subtitling mode with the tabs at the top of the editor. Each cue is coloured red or green against the formatting rules; the three dots next to Subtitles, then Edit rules, takes the client's characters per line, lines on screen and cue length.
• Type to fix a word, Enter to split a cue that runs long, merge cues to join two short ones, drag the handles or type an exact timestamp when a cue starts late. Every cue green: export, SRT or VTT.

Path B: one request to OpenAI's transcription endpoint

• response_format on /v1/audio/transcriptions is one of json, text, srt, verbose_json, vtt or diarized_json. For gpt-4o-transcribe and gpt-4o-mini-transcribe the only supported format is json; gpt-4o-transcribe-diarize returns json, text or diarized_json. The subtitle model is whisper-1: model=whisper-1, response_format=srt, and the response is the cue file.
• Files are capped at 25 MB: extract the audio track and keep it compressed; split anything larger at a pause, not mid-sentence. whisper-1's prompt takes up to 224 tokens, which is where the names and acronyms go.
• English only, from a foreign-language video: the separate /v1/audio/translations endpoint does it in one step, whisper-1 only.

The fix pass before translation

• Play the video against the cues once. The things that break are the same every time: a name spelled by ear, a number written as words, a cue that starts a second late, a cue that holds two sentences.
• In the editor: a typed correction, a split with Enter, a dragged handle. On the API route: edits in any text editor, since SRT is plain text with numbered cues and 00:01:23,400 --> 00:01:26,900 timecodes.

Translating the cue file with the timecodes locked

• Paste the whole SRT into a text model with a short instruction: translate only the text lines, keep every cue number and timecode exactly, do not merge or split cues, keep each line under the client's limit.
• Ten minutes of speech is roughly 1,500 words; 100 tokens are approximately 75 words in English, with other languages at different ratios. Call it about 2,000 tokens in and 2,000 out per language.
• gpt-4o-mini at $0.15 in and $0.60 out per million: a fraction of a cent per language; gpt-5.6-luna $0.20 and $1.20. Mistral Large $0.5 and $1.5, batch at half price. Gemini 3.8 Flash $0.75 and $3.75 on the Paid tier through December 31, 2026, and the tier matters more than the cents: Free tier content is used to improve Google's products, Paid is not, and a client's video should not go through the Free door. The drafting page reads all four lists in full.
• Read each translated file against the video once more. A cue that fit in English can fail the line rule in German or Vietnamese; the fix is a shorter phrasing or a split cue. Export, name by language code, deliver all three with the source.

If the client wants the voice too

• ElevenLabs' Dubbing page: upload the video or paste its URL, choose the languages, adjust Speaker similarity, generate. The cost is shown for confirmation before the dub starts, charged per minute of source media for each language; files up to 2 GB and 180 minutes. Dubbing v2 is automatic, with no in-app editing; editing via the API is Enterprise only.
• 2,000 credits a minute automatic with a watermark, 3,000 without: ten minutes into two languages is 40,000 or 60,000 credits, more than a Starter month of 30,000 and half a Creator month of 121,000 at $22.
• Free-plan dubs are watermarked with no option to remove it, on top of the free plan's rule that its output cannot be used for any commercial purpose. The voiceover page reads that licence line in full.

The minute on two price lists

The chart above puts one minute of video on all three lists. The machine steps: transcription at $0.003 a minute on gpt-4o-mini-transcribe or $0.006 on Whisper, the model that returns SRT, or $0.22 an hour on ElevenLabs' Scribe v2 API; a translation step at a fraction of a cent per language; $0.034 a minute if you want OpenAI's live translation instead. The human service: ElevenLabs' Productions team from $2.20 per minute, with its language teams translating if you choose. The person: Rev pays captioners $0.54 to $1.10 per minute. The AI job cost calculator runs the machine rows for any length.

Who owns the captions, and where the audio goes

• OpenAI assigns to the customer all of its right, title and interest in the Output; the customer keeps the Input; as of March 1, 2023 data sent to the API is not used to train its models unless you opt in; all Customer Content is deleted within thirty days of termination. The SRT files on your disk are the only copies.
• Gemini's rule is the tier row above. ElevenLabs' free plan has no commercial licence; every paid plan does.

If you are the person doing step three

Rev pays weekly via PayPal for all approved work, within 1 week of approval, and applications sit on a waitlist due to high application volume. The rate is per minute of video, not per minute worked, so the wage is $0.54 to $1.10 divided by the minutes you spend on each minute of video. Nobody on this page publishes that division.

What to do today

1. Take a ten-minute video you have rights to. Run it once through ElevenLabs with Include subtitles on and once through whisper-1 with response_format=srt, and time the fix pass on each.
2. Translate one into one language with the timecodes locked and read it against the line-length rule.
3. The machine bill will be a few cents; the fix pass is your cost, and $2.20 a minute is the price of the human alternative the client can also buy.