THE SHORT ANSWER
THE SHORT VERSION
- One creator profile (a creator_id, plus the network's own external_id) carries that creator's Do Not Translate list and context brief onto every dub and live session.
- Entries attach at three scopes, creator, stream, or job, and the narrower scope wins.
- The dubbed voice comes from the creator's own clip: Familiar sounds 27.8% more like the real speaker than ElevenLabs Dubbing v1 across 418 paired outputs (August 5, 2026).
- Every language ships as a finished video, .srt and .vtt subtitles, two audio files, and a translated title and description with links and hashtags untouched.
01.Why does one workflow per creator break?
A multi-channel network (MCN) dubbing 40 creators into six languages runs 240 dubbed channels, and each creator brings their own names, catchphrases, running jokes, and sponsor codes.
- Where a per-creator process fails. A glossary kept in a document is applied by whoever reads it, and a change to one creator's list reaches the next dub only if someone carries it over.
- The audience it costs. 64 of the 100 biggest YouTube channels aren't in English.
02.How do per-creator settings ride one integration?
The network integrates once under its organization's API key; each creator is a profile the pipeline reads on every request.
- One profile per creator. POST /v1/creators returns a creator_id; the network stores its own external_id on the profile and looks creators up by it (Creator profiles).
- The list and the brief follow the profile. A dub or live session tagged with creator_id gets that creator's Do Not Translate list and context, matched case-insensitively across the dub, the subtitles, and the translated metadata (Do Not Translate).
- Three scopes, narrower wins. Entries attach at creator (persistent defaults), stream (saved with the live session, reused next run), or job (one upload); a job entry overrides a creator default for that video only.
- Seeded defaults, one sentence of context. New profiles start with platform names and common gaming and internet terms protected; a context brief like "A chess channel: openings, blunders, and speedruns" keeps a name a name (Context).
- One list, one webhook stream. GET /v1/dubs filters by creator_id and status; every state change arrives at one endpoint, signed and retried for 24 hours (Webhooks).
| SETTING | SHARED BY THE NETWORK | PER CREATOR |
|---|---|---|
| API key | One organization key on every request | None: creators are profiles under that key; there is no creator login |
| Do Not Translate | Seeded defaults: platform names, common gaming and internet terms | The creator's own entries, keyed by creator_id; a job entry overrides for one video |
| Context brief | None | One sentence per creator; a stream or job sentence overrides it |
| Target languages | Up to 29 per dub | Chosen per request |
| Review step | review=auto renders straight through; review=manual pauses at in_review | Chosen per dub |
| Webhooks | One default endpoint, signed, retried for 24 hours | webhook_url override per dub or live session |
| Dub list | GET /v1/dubs across the organization | creator_id filter narrows it to one creator |
| Default limits | 100 requests a minute, 10 concurrent renders, 2 live sessions, 10 GB per file, 24 h of source | None per creator; raised by the organization's point of contact |
03.Does the creator still sound like themselves?
The voice in the dub comes from the creator's own clip, and the whole face performs the new language: eyes, brow, cheeks, head. A stock narrator would sell the audience a different person.
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 (0.4605 vs 0.3604); all 11 languages favored Familiar.
ELEVENLABS DUBBING V1 · 418 PAIRED OUTPUTS · 11 LANGUAGES · AUGUST 5, 2026- Vs YouTube auto-dubbing, a separate study. On a frozen cohort of 50 paired dubs (31 aligned; Arabic, Spanish, French, Hindi, Korean; captured August 8, 2026), Familiar measured 182.9% higher speaker resemblance (0.399 vs 0.141) (the study).
04.Do livestreams fit the same shape?
A live session tagged with creator_id applies the same list and context, and the context sent with the session writes the stream scope. Live sessions dub from 11 source languages into the other 29, 5 to 9 seconds behind the original, everyone on camera in their own voice; the broadcast comes in over WHIP (WebRTC), RTMP, or SRT, and each language's feed pushes to its own channel. Live has no review step; the list and the brief are the controls (Real-Time Dubbing for Livestreams, Live streaming).
05.What does the network get to post, per language?
- The finished video (mp4), downloadable the moment dub.output_ready fires for that language (Outputs & subtitles).
- Subtitles in .srt and .vtt, free on every dub, from the reviewed transcript.
- Two audio files. The full mix (m4a) for platforms that accept additional audio tracks; the voice alone (wav) for the creator's own mix.
- The translated title and description, with links, @handles, #hashtags, promo codes, and chapter timestamps left exactly as written (Metadata translation).
- Review, chosen per dub. review=manual pauses at in_review so any line is edited before render; review=auto, the default, renders straight through.
06.What should the network test before it commits?
- Pick the hardest creators. Two or three with the most names, sponsor codes, and running jokes; load their lists first, then dub paired clips (How to Run an AI Dubbing Pilot).
- Count corrections at the review step. With review=manual, the lines a native reviewer changes are the measurement; any phrase they protect goes onto the creator's list for every later dub.
- Check the roster's formats. Clean, stable footage dubs best: podcasts and interviews with 2 or 3 people on camera, lectures; everyone on camera gets dubbed.
QUESTIONS
How does a network dub many creators without a separate workflow for each one?
One profile per creator holds that creator's Do Not Translate list and context brief. Every dub request and live session carries the profile's creator_id, so the network runs one integration, lists dubs by creator, and receives every state change on one webhook endpoint.
Can a creator keep their catchphrases and inside jokes?
Yes. The creator's Do Not Translate list keeps catchphrases, names, and signature slang exactly as said in the dub, the subtitles, and the translated title and description. Jokes are re-landed in the target language, and the context brief keeps a running joke's setup.
Do creators sound like themselves in the dub?
The voice comes from the creator's own clip. In the ElevenLabs Dubbing v1 study (418 paired outputs, 11 languages, August 5, 2026), Familiar sounds 27.8% more like the real speaker, and all 11 languages favored Familiar.
Does each creator need their own Familiar account?
No. Creators are profiles under the network's integration: POST /v1/creators takes a display name and the network's own external_id and returns a creator_id to pass on dubs and sessions. Who may open which settings screen is decided in the network's own product.
REFERENCES
- [1]Creator profiles: creator_id, external_id, and the three scopes (API reference)
- [2]Do Not Translate: matching, scopes, seeded defaults (API reference)
- [3]Context: the one-sentence brief per creator, stream, or video (API reference)
- [4]Metadata translation: what survives untouched (API reference)
- [5]Familiar vs YouTube auto-dubbing: the frozen 50-pair cohort, full data
- [6]Familiar vs ElevenLabs: both paired studies, intervals, listening examples
Dubbing is finally good. See the measurements, then talk with the team about your catalog.
FAMILIAR · THE LAUNCH FILM · 2:31