Create project
Create a dubbing project from an uploaded file (file) or a source URL (source_url).
Returns as soon as the project record exists, before the source has been fetched: the project starts queued and reaches ready once its source has been transcribed. Add a language target to it for each language you want, or pass target_language to queue the first one here.
Creating a project incurs a minimum charge of one language: you are charged for at least one language’s dubbing when the project is created, before any output exists. This charge prepays the first language target — the first target you add, or the one queued via target_language, consumes it — and each additional language is charged separately.
Preparation can take minutes on a long source, so we recommend passing webhook_ids to be notified when the project turns ready or failed, rather than polling for it.
Headers
Request
The source media file to dub: an audio or video file of at most 3 GiB. Provide this or source_url, not both.
Public HTTP(S) URL the source media is fetched from server-side, subject to the same size and format limits as an upload. Provide this or file, not both.
Optional free-form string (at most 500 characters) to identify the project on your end. Stored and echoed back verbatim; it does not affect the dub.
BCP-47 language tag of the source media; must be a language the transcription model supports. Any region or script subtag is ignored, since transcription is per-language. Omit to auto-detect.
Dubbing model (dubbing_v1 or dubbing_v2) every language target of this project is dubbed with. Defaults to dubbing_v2. Fixed at create time — the source is prepared for this model, so neither the project nor an individual target can change it later.
Key terms to bias transcription and translation toward (for example, product or brand names). At most 1,000 terms; each term at most 50 characters and 5 words; the characters <>{}[]\ are not allowed. Terms are trimmed and deduplicated. Repeat the field once per term, or pass a single JSON array.
IDs of workspace webhooks to notify as this project progresses — the alternative to polling, and what we recommend. Each receives a dubbing_project_ready or dubbing_project_failed event for the project, and a dubbing_language_completed or dubbing_language_failed event for every language under it; dubbing_language_completed carries the output download URLs. At most 3 IDs, each already configured in your workspace — see Webhooks for how to create one and verify its signature. Repeat the field once per ID, or pass a single JSON array or comma-separated string. Delivery is best-effort and can repeat, so we recommend handling events idempotently.
Optional shortcut: also create a language target in this BCP-47 language, queued to start once the project is ready — equivalent to creating the project and then creating one language target. Must be one of the languages the dubbing model supports, and a region-qualified tag must be one of the supported dialects. This is the first language target, so it consumes the project’s minimum charge rather than adding to it. Its ID is returned in language_ids.
Enterprise only. Optional JSON transcript to use instead of transcribing the source: a {"segments": [...]} document, at most 20,000 segments and 4 MiB. See Bring your own transcript for the segment fields and their constraints. source_language is required whenever a transcript is provided. If any segment carries a translation, target_language is required and every segment must carry one; those translations seed the target created via target_language, which then skips machine translation.
Response
Lifecycle status of the project: queued before the source is picked up, preparing while it is transcribed, ready once transcription is done and language targets can start, or failed. A project is never reported as processing — that value belongs to language targets.
Monotonic counter incremented whenever the source transcript is edited (segment add/edit/delete).
The free-form string you supplied as reference when creating the project, or null if you supplied none.
BCP-47 language tag of the source media (null if auto-detected).
Source media metadata, populated once the source has been fetched and decoded (shortly after create, before the project is ready); null until then.
Identifiers of the language targets under this project. Populated when a single project is fetched, and on create when target_language creates one. Always empty in list responses — list the project’s language targets instead.
IDs of the workspace webhooks notified as this project and its languages reach ready, completed, or failed.
Why the project failed; null unless status is failed. Also null for the few projects that failed before failure reporting was introduced.
Non-fatal conditions raised while preparing the source, empty when there are none. Reflects the latest preparation. Conditions raised while dubbing a particular language are reported on that language instead.