Upload a title, pick the languages, download the cue files. Word-level timing, reading-speed limits enforced on every cue, and right-to-left scripts that render the right way round.
They said the bridge would stay closed until Monday.
הם אמרו שהגשר ייסגר עד יום שני.
Dijeron que el puente estaría cerrado hasta el lunes.
उन्होंने कहा कि पुल सोमवार तक बंद रहेगा।
قالوا إن الجسر سيبقى مغلقاً حتى يوم الاثنين.
Ils ont dit que le pont resterait fermé jusqu'à lundi.
No implementation project, no onboarding call, no sales engineer. Sign up, load the wallet, drop in a file. The dashboard shows exactly what each job cost and what came back.
Flagged cues exceeded the reading-speed limit and are worth a look before publishing. Everything else conformed cleanly.
Left: the project list, with the wallet balance and every job's real cost. Right: a finished job, with cue counts and flag counts per language before you download anything.
You own the rights. The file is encoded and sitting on your origin. The only thing between it and an audience in another country is a text file with timecodes in it.
Subtitling houses quote per minute and take a week. For a modest catalogue in a few languages that is a five-figure invoice on a schedule measured in months — so what usually happens is that nobody does it at all.
Typical market rates per hour of runtime, per language. The other rows are indicative, not quotes — get your own. Bars use a logarithmic scale — the real gaps are wider than they look.
The longest part is waiting, and you can close the tab while it runs. Nothing to install, nothing to configure.
Step 01
Top up from $100.00, which covers 25 hour-long titles in a single language. Credit never expires, there is no plan to cancel, and an idle month costs nothing at all.
Step 02
Large files go up in 5 MB pieces, so a multi-gigabyte master arrives reliably even on a shaky connection. Your browser reads the runtime straight from the file, so the exact cost appears before you confirm anything.
Step 03
Up to 8 per job. Each is a separate translation pass, but transcription only happens once — which is why the fifth language costs us less than the first.
Step 04
WebVTT, SRT and a plain transcript for every language. Drop them into your player, add them to an HLS manifest, or hand them to a distributor.
Move the sliders. The figure on the right is the whole bill — no setup fee, no per-format charge, no minimum commitment, no tier to model.
Features usually run 90 to 140 minutes. Episodes, 22 to 50. Anything under an hour bills the flat $4.00.
Each language is a separate translation pass, so each one bills separately.
Backfilling a library? Set this to how many you want to clear each month.
Per month
$56.00
House comparison uses $150 per hour per language — the low end of a subtitling vendor quote.
Wiring a speech API to a translation model takes a weekend and produces a file. It does not produce subtitles. These six failures are the difference.
01
A cue above roughly 17 characters per second cannot be read before it vanishes. Raw machine output routinely doubles that.
02
Hebrew, Arabic, Hindi and Spanish all inflect for who is speaking and who is spoken to. Line-by-line translation gets this wrong about half the time by chance.
03
A Hebrew line with a Latin brand name renders in the wrong order without cue direction settings. Not gibberish — confidently wrong.
04
Breaks belong at clause boundaries, not wherever the character count landed. Bad breaks measurably slow comprehension.
05
Under a second reads as a flash; over seven reads as a frozen track. Long speeches need splitting, rapid exchanges merging.
06
Music, crowds and overlapping dialogue degrade recognition. We surface which cues are shaky rather than presenting every line with equal certainty.
One translated line of dialogue and the 6.4 seconds it has to live in. Drag the reading-speed limit and the conformance engine re-cuts it live — this runs the same splitting and line-breaking logic the server does.
Red cues exceed the limit even after splitting — those get flagged rather than quietly shipped. Broadcast work usually sets this to 15; streaming platforms typically accept 17.
A Hebrew subtitle containing a Latin brand name and a time. Same text, same player. The only difference is whether the cue carries direction settings.
The Latin run and the digits get reordered against the Hebrew. A reader sees a sentence that is subtly, confidently wrong — worse than obvious garbage, because nobody catches it in QA.
Hebrew flows right to left, the embedded Latin run flows left to right inside it, and the time stays intact. Our WebVTT writes the direction settings; the SRT writer adds a UTF-8 BOM and RLM marks, because SRT has no encoding declaration at all.
Rendered by your own browser's bidirectional engine — the same one your player uses. Nothing here is faked.
Coverage is the easy claim — anyone can list a hundred. What matters is whether the hard scripts render correctly in your player.
Amber pills are right-to-left scripts and get the direction handling shown above. Source audio can be in any of these plus roughly sixty more we recognise but do not currently translate into.
Every format is included, and re-exporting a title you have already processed costs nothing — you paid for the translation, not the serialisation.
WEBVTT 1 00:04:12.480 --> 00:04:15.920 direction:rtl align:right הם אמרו שהגשר ייסגר עד יום שני. 2 00:04:16.010 --> 00:04:19.640 direction:rtl align:right אף אחד מהכפרים בצפון לא יכול היה להגיע לשוק.
(UTF-8 BOM, then RLM-prefixed lines)
1
00:04:12,480 --> 00:04:15,920
הם אמרו שהגשר ייסגר
עד יום שני.
2
00:04:16,010 --> 00:04:19,640
אף אחד מהכפרים בצפון
לא יכול היה להגיע לשוק.
<!-- plain HTML5 player --> <video controls src="film.mp4"> <track kind="subtitles" src="film.he.vtt" srclang="he" label="עברית" default> <track kind="subtitles" src="film.ar.vtt" srclang="ar" label="العربية"> </video> # HLS master playlist #EXT-X-MEDIA:TYPE=SUBTITLES,GROUP-ID="subs",\ NAME="Hebrew",LANGUAGE="he",URI="he.m3u8"
Files stay in your account permanently. Delete a job and they go with it; otherwise re-download any time without paying again.
Transcription runs once per title no matter how many languages you order. Only translation repeats — which is why we can keep one flat rate instead of a tier ladder.
What a two-hour feature costs as languages are added.
40-title catalogue
Forty 90-minute titles in four languages. Once — then only new releases each month.
Weekly series
45 minutes × four episodes in three languages. Per month, every month.
One festival submission
A 112-minute film, English only. Less than the courier fee for the disc.
Your subscriber base is more multilingual than your content. Work through the back catalogue a batch at a time and add languages as demand shows up.
Every territory deal needs local-language deliverables. Turn a two-week subtitling schedule into an afternoon and stop letting localisation set your release date.
Festivals want subtitles and there is no post budget left. Eight dollars is not a budget line.
Lecture libraries are the highest-volume, lowest-margin subtitling job there is — exactly the shape this handles well.
Small budgets and languages the big vendors treat as an afterthought. Hebrew, Punjabi and Tamil are first class here.
It depends far more on the audio than on the language. A clean interview or lecture lands in the high nineties on word accuracy. A feature with a score under the dialogue, crowd scenes and overlapping speech is meaningfully worse — which is exactly why every cue carries a confidence score and the flag count is shown per language. Run one title for a few dollars and you will have real numbers for your own material instead of ours.
For most catalogue work, no. For flagship titles, anything with legal exposure, or comedy where the wordplay carries the scene, yes. The practical approach is to review only the flagged cues — usually ten to fifteen percent of the file — which turns a week of work into an afternoon.
MP4, MKV, MOV, MXF, TS, AVI, WebM and the common audio formats, up to 2 GB. Uploads are sent in 5 MB chunks, so file size and a shaky connection are not a problem — a dropped transfer resumes from the last confirmed piece rather than starting over.
The media file is deleted as soon as the transcript comes back — normally within minutes of upload, not days. Only the text survives. Subtitle files stay in your account so you can re-download them, and you can delete those whenever you like.
Not here. Dubbing costs roughly a hundred times what subtitles cost per hour and the speaker's lips still move in the original language. For almost every catalogue subtitles are the better spend, and we would rather say that than sell you the expensive thing.
No trial credit. The minimum top-up is $100.00, which covers 25 hour-long titles in a single language — enough to put a real film and several episodes through and judge the output on your own material rather than on a demo file. Credit never expires, so nothing is wasted if you take your time.
You are not charged. The hold on your wallet is released automatically and the ledger shows the refund. If a job completes but the output is genuinely below what the source should have produced, tell us and we credit it back.
Create an account, top up the wallet, and process a title you actually care about. A feature film in two languages costs about $16.00, so you can judge the output on real material rather than on a demo clip.