THE SHORT ANSWER
THE SHORT VERSION
- One dub request takes a title into up to 29 target languages; the manual review flag holds each language at a review step until a reviewer approves it.
- Each language renders on its own approval, so Spanish can publish while Japanese is still under review.
- Webhooks report every state change: a dub ready for review, each language's output ready, the whole dub complete, or a failure with its error.
- Measured on the same clips on August 5, 2026, ElevenLabs Dubbing v2 (Alpha) had 270% more background-sound error: the laughter, the music, the ambience.
01.Why is coverage the constraint?
A catalog adds titles every week; a pipeline that casts, records, and mixes one title at a time adds languages at a fixed rate, so the queue sets how much of the catalog a market can watch in its own language.
of the 100 biggest YouTube channels aren't in English.
- Full human dubbing runs $70 to $150 per finished minute, translation plus voice: $4,200 to $9,000 for a 60-minute title in one language (How Much Does Dubbing Cost in 2026?).
- Every language is its own queue in a per-title pipeline: ten languages for one title means ten casting, recording, and mixing jobs, and the slowest one sets the release date.
- Audio-only dubbing leaves the mouth speaking the source language; viewers read lips that contradict the words they hear (Why the Dub Feels Off Abroad, and How Studios Fix It).
02.What does dubbing on upload look like?
One integration covers the loop: the upload flow makes one request, the API reports back at every step, and the finished files serve from the platform's own player.
| STEP | WHAT HAPPENS | WHAT THE PLATFORM RECEIVES |
|---|---|---|
| Upload | POST /v1/dubs with the file or a source URL, up to 29 target language codes, and review=manual; the source language is detected or pinned. | dub_id, status queued, one outputs[] entry per language |
| Transcribe and translate | Speech-to-text with word timestamps, speakers separated; every line translated with the Do Not Translate list and the context brief applied. | status transcribing, then translating |
| Review | The dub pauses at in_review. GET the transcript per language, PATCH any line, POST /render for one language or all of them. | dub.ready_for_review, then the editable transcript |
| Render | Voice and lip-sync generated together, per language; each language carries its own status. | dub.output_ready per language; dub.completed once every language is rendered |
| Deliver | One output per language: the finished mp4, .srt and .vtt subtitles from the reviewed transcript, the full audio mix or the voice track alone, and the translated title and description. | GET /v1/dubs/{dub_id}/output/{lang} with format=mp4, srt, vtt, audio, or voice |
- Failures are events too. dub.failed carries the language and the error; a delivery not acknowledged within 10 seconds is retried for 24 hours, and duplicates are deduped on the event id (Webhooks).
- Names and catchphrases hold. A Do Not Translate list keeps them exactly as said; a one-sentence context brief (who is on screen, what the show is) sharpens every line; both scope to a creator profile, a stream, or a single title (Do Not Translate, Context).
- Metadata travels with the title. The translated title and description arrive per language; links, @handles, #hashtags, promo codes, and chapters survive untouched (Metadata translation).
03.Why does one model change the coverage math?
Stitched pipelines trade throughput against quality: every extra tool is another pass and another bill, and two models can't hold one identity. One unified model generates the voice, the lips, and the scene together.
ElevenLabs had 270% more background-sound error: the laughter, the music, the ambience.
11.92 VS 3.22 DB · ELEVENLABS DUBBING V2 (ALPHA) · 111 PAIRED OUTPUTS · MANDARIN, SPANISH, JAPANESE · AUGUST 5, 2026Familiar sounds 27.8% more like the real speaker; all 11 languages favored Familiar.
0.4605 VS 0.3604 · ELEVENLABS DUBBING V1 · 418 PAIRED OUTPUTS · 11 LANGUAGES · AUGUST 5, 2026Familiar made 54.7% fewer important spoken translation errors (29 vs 64).
ELEVENLABS DUBBING V2 (ALPHA) · THE SAME 111 PAIRED OUTPUTS AS THE 270% TILE · AUGUST 5, 2026- Translation quality holds. Familiar's translations scored 100.3% of the first-pass expert professional's quality, averaged across French, Chinese, Hindi, and Indonesian, scored blind against expert reference edits (August 6, 2026): a match (the study).
- The scene survives the dub. Music is never dubbed; a song passes through untouched, and the laughter, sound effects, and room ambience stay under the new speech; every speaker on camera is dubbed in their own voice.
04.Which titles first?
Per-language release makes the planning unit a title in a language: a market goes live with the languages that have cleared (Multi-Country Launches: Dubbing Every Market at Once).
| TIER | LANGUAGES |
|---|---|
| Tier 1 (14) | Cantonese, English, French, German, Indonesian, Italian, Japanese, Korean, Mandarin, Portuguese, Russian, Spanish, Thai, Vietnamese |
| Tier 2 (9) | Bulgarian, Catalan, Croatian, Dutch, Greek, Lithuanian, Norwegian, Slovak, Swedish |
| Tier 3 (7) | Arabic, Belarusian, Danish, Kazakh, Latvian, Serbian, Ukrainian |
- Cleanest formats first. Podcasts and interviews with 2 or 3 people on camera, straight-to-camera video, lectures, presentations, and clean single-camera scenes.
- Tier 1 first, then widen. Every one of the 30 languages dubs from and into every other, so the tier order is a rollout order (Languages).
- Route by risk. review=manual holds a title at in_review until a market reviewer renders it; review=auto (the default) renders straight through; subtitle jobs always stop, since their text comes from the review step.
- Broadcasts are a separate lane. Live sessions dub a broadcast 5 to 9 seconds behind the original over SRT, RTMP, or WHIP, from 11 source languages into the other 29, each language's feed pushed to its own destination; live has no review step (Real-Time Dubbing for Livestreams).
05.How do you pilot it on your own catalog?
- Pick the hardest title. One representative episode or a connected run in two or three priority languages, with fast dialogue, a joke, music under speech, proper names, and one late edit.
- Write the acceptance criteria before anyone listens. The words, the voice, the scene, the face, and the timing, each scored by a native reviewer; plus revision cycles and time from source change to approved replacement.
- Count the corrections. The lines a reviewer changes at the review step, by number and kind, are a direct quality measure; every change ships in the dub and its subtitles.
Paired design, the scorecard, and how to read the results: How to Run an AI Dubbing Pilot (2026). Access is by Studio contract, sized to the catalog, with 30 day money back on signed annual contracts.
QUESTIONS
What are streaming services using to add local-language dubs faster?
Dubbing on upload through an API: one request per title into up to 29 target languages, a review step per language, webhooks reporting each state change, and the finished video, subtitles, audio tracks, and translated title and description served from the platform's own player.
Does faster dubbing mean worse dubbing?
Not when the voice, the lip-sync, and the scene come from one model. On the same clips (August 5, 2026), ElevenLabs Dubbing v2 (Alpha) had 270% more background-sound error than Familiar. Familiar's translations scored 100.3% of first-pass expert professional quality across French, Chinese, Hindi, and Indonesian: a match.
Can we release one language before the others?
Yes. With review=manual each language holds at in_review and renders on its own approval: POST /render with a langs list releases Spanish today and the rest after their reviewers sign off; each language fires its own dub.output_ready webhook when its files are downloadable.
What does full human dubbing cost, for comparison?
Full human dubbing, translation plus voice, runs $70 to $150 per finished minute: $4,200 to $9,000 for a 60-minute title in one language. Translation alone runs $22 to $45 per spoken minute at published per-word rates.
Which languages can a platform dub into?
30 languages, any to any: a title in any of the 30 dubs into the other 29, in three published quality tiers, best first. Tier 1 holds 14 languages, Tier 2 nine, Tier 3 seven, alphabetical inside each tier.
REFERENCES
- [1]The Familiar API: introduction, quickstart, and the dub lifecycle
- [2]Transcript & review: hold, edit, and render per language
- [3]Webhooks: the event catalog, delivery, and retries
- [4]Familiar dubbing benchmarks: three paired studies, full data and listening examples
- [5]How Much Does Dubbing Cost in 2026? (every published dubbing rate, per finished minute)
- [6]ElevenLabs Dubbing documentation: billed per minute of source media, per target language (accessed September 6, 2026)
Dubbing is finally good. See the measurements, then talk with the team about your catalog.
FAMILIAR · THE LAUNCH FILM · 2:31