Professional dubbing, directed by you.
Picture locked · Sound in progress
Professional dubbing, directed by you.
WooblaDub is a dubbing workspace. You rebuild the dialogue and nothing else — the picture is never re-rendered, and the music, effects and ambience of the original scene stay underneath it.
Start in the StudioSee the workflow
Picture · untouched
\--:--:--:--
00:0000:0200:0400:0600:08
Dialogue · ENRebuilt
ROLE AROLE BSEO-YEONJUN-HO
Music & effectsPreserved
Ambience · room tonePreserved
Workspace diagram · one track is replaced, the rest of the scene is not
- Broadcasters
- Feature & episodic
- Vertical drama
- Independent film
- Distributors
- Localization teams
- Editor and revision history
- Production files included
- Storage billed separately
- Shared casts and glossaries
- Team review and approvals
- Priority processing
- DAW-ready interchange
§ 01 — The position
EVALUATION CLIP · TEARS OF STEEL · EN → DEBuffering both versions…00:00:00:00
OriginalWooblaDub
▶Play the evaluation clip
PlayOriginal production dialogue, as delivered.
Action-sequence performerSource · EN → Adaptation · DE
Nooooo!
Neeein!
Directionsustained panic, collapsing at the end
All 6 lines 1. Action-sequence performerNooooo!Neeein! 2. CeliaYou broke my heart.Du hast mir das Herz gebrochen! 3. ThomI know.Ich weiß. 4. CeliaYou forgot me on earth.Du hast mich auf der Erde vergessen. 5. ThomI know.Ich weiß. 6. CeliaI should just crush you.Ich sollte dich einfach zermalmen!
Thirteen seconds of Tears of Steel, English into German. Six lines, two characters, one argument. Music, effects and room tone are the original recording in both versions; only the dialogue was rebuilt.
Dialogue synthesised · AI-generated audioTears of Steel · (CC) Blender Foundation | mango.blender.org · CC BY 3.0
A dub is not a translation. It is a second performance, and someone has to direct it.
This is a working environment, not a button. You cast the voices, shape the reading and adjust the line against picture. The machine does the labour; the judgement stays with you. We are in evaluation with launch partners, and no language route has passed a locked-media review yet — the Languages page says which ones are close.
We are
A multilingual dubbing production workspace. Dialogue is rendered through a voice provider account you connect — a licensed stock voice, or one you hold the rights to. The adaptation, casting, direction and approval stay with you. Picture-immutable by default.
We are not
An instant video translator. A face-regeneration product. A voice-model owner. An unattended one-click output.
Locked cut to delivery, in six passes.
§ 02 — The workflow
01
Bring in the cut
Video, screenplay, SRT or transcript. The timeline is built from what you already have.
02
Adapt the dialogue
Translate, then rewrite for mouth and rhythm. Source, gloss and adaptation on one screen.
03
Cast the voices
One voice per character across a whole title. Casts bind to characters, so they survive a season.
04
Direct the take
Quieter. Further back in the room. Slower. You write the note a director would give.
05
Review and approve
A-B the original against the dub anywhere on the timeline. Every decision keeps its history.
06
Deliver the record
Final video, stems, cue sheets, subtitles and the QA record a distributor asks for.
Three things get called dubbing.
§ 03 — The comparison
| How a human-directed dubbing workspace differs from an automated video translator and from a visual dubbing product| | Automated translator | Visual dubbing | WooblaDub | | -------------------------------------------------------------------------------------------------------------------- | -------------------------------- | ----------------------------------- | ----------------------------------- | | The picture | Untouched | Re-rendered to fit the new dialogue | Untouched, and checksummed | | Music, effects, room tone | Usually flattened into one track | Preserved | Preserved, provenance recorded | | Changing one line | The whole output regenerates | The whole output regenerates | That cue reruns | | How you direct it | A prompt | A prompt | A note, cue by cue | | Whose voice models | The vendor's | The vendor's | Yours, in your own account | | What you receive | A finished file | A finished file | Stems, script, cue sheet, QA record |
The performer decides. The editor is accountable.
§ 04 — Consent & rights
There is no mechanism in WooblaDub that would let us own, resell or clone a voice, because we never hold one.
We do not own voice models
You connect your own provider account. Consent stays inside the agreement you already hold.
We do not train on your material
Your material produces your dub. It does not improve output for anyone else.
No performer is synthesised without agreement
Where a provider holds a performer’s consent, that provider stays the record of it.
An editor of record signs the work
Every approval carries a name and a date. Since 2 August 2026 the EU AI Act has expected someone to be identifiable as responsible for a synthesised performance. That is the same person.
§ 05 — The production record
Your project leaves with its production files.
No black box. Everything the dub was made of leaves with you, in the formats a mixer and a QC house already use.
EVALUATION CLIP · TEARS OF STEEL · EN → ESBuffering both versions…00:00:00:00
OriginalWooblaDub
▶Play the evaluation clip
PlayOriginal production dialogue, as delivered.
A different scene, English into Spanish. Every element you can hear other than the dialogue shipped in the package as its own stem.
Dialogue synthesised · AI-generated audioTears of Steel · (CC) Blender Foundation | mango.blender.org · CC BY 3.0
Final video — dubbedMP4 / H.264
Dialogue stemWAV 48 kHz
M&E stem — music and effectsWAV 48 kHz
Adapted scriptJSON
Cue sheet & timingsCSV + CMX3600 EDL
SubtitlesSRT / WEBVTT
Casting & voice mapJSON
Revision & QA recordJSON + JSONL
KO / ZH → ENIn trial
Launch focus · film, episodic and vertical drama
ES / DE / IT → ENIn trial
Launch focus · European film and television
HI → ENIn trial
Launch focus · feature and episodic
Any other pairOn request
Available after language and voice review
How a route becomes qualified →
Priced per project, from one scene to a full title.
§ 06 — Access
Pay as you go
Per project
One scene, one episode or a full title. No subscription.
Pay as you go Pro
Priority production
Faster queues, series continuity, deeper delivery controls.
§ 07 — Straight answers
The questions that come up first.
Access opens in stages. Create an account, tell us what you are dubbing, and we will set the room up around it.
Start in the StudioWhat moves a quote
Whose voices are these?
Yours to choose. Dialogue is rendered through a speech provider account you connect, so the voice is either a licensed one from that provider's pool or one you hold the rights to and bring yourself. WooblaDub never owns or holds a voice model. What is human in every case is the adaptation, the casting, the direction of each reading and the approval — each carrying the name of the person who made it.
Do you change the actors' faces, and how does lip sync work without that?
No, and by writing to the mouth rather than changing it. Adaptation is timed against the source speaking window and the visible articulation, which is the constraint a dubbing adapter has always worked inside. Source and output video stream hashes ship with every delivery, so that the picture was untouched is checkable rather than promised. Products that reanimate the performer's face solve it the other way, and on a tight close-up they solve it better — that is a different product with different consent questions.
Whose voice is being cloned?
None, unless you bring one and the rights to it. The default is a stock voice from your provider's licensed pool. We do not own or sell voice models, and we will not synthesise a performer who has not agreed to it — the digital-replica terms your production is bound by are the terms this workflow runs inside.
What happens when I change one line?
Only the affected cue reruns, along with whatever depends on it. The revision graph shows what changed and what stayed.
Is the M&E reconstructed?
If you supply one, it passes through untouched and is recorded as supplied. If you don't, we separate one and label it a reconstruction, and the reconstruction has to prove itself: the separated dialogue and background must sum back to the original mix, channel by channel, above a published fidelity threshold. That is not the same thing as a filled M&E: where footsteps and cloth are buried in the dialogue, recovering them is a foley session, not a processing step.