Voice cloning request handling

ElevenLabs stores each voice clone request, the source audio, consent status, and the resulting voice model, under a project. A Beam agent checks a new clone request against the account's approved consent policy, submits it for processing, and updates the project record once ElevenLabs returns the finished voice model. Requests missing a consent file, or using a voice similar to one already flagged, are held for a person to clear before the voice model is generated. This keeps routine cloning moving while anything touching rights or consent still gets a human decision before it proceeds further.

Long-form narration rendering

ElevenLabs turns a script into long-form narration and records the render status, duration, and output file against the project. A Beam agent watches the render queue, downloads a finished audio file once ElevenLabs marks it complete, and attaches it to the matching content record so the writer doesn't fetch it manually. A failed render, or output well outside the expected length for that script, is flagged for a person to check before it reaches a listener. Scripts containing claims the account hasn't pre-approved for narration are also held rather than sent straight to render.

Automatic dubbing turnaround

ElevenLabs's dubbing tool maps a source video to a target language track and records the job status per asset. A Beam agent submits an approved video for dubbing, checks back on job status, and updates the asset record with the finished track once it's ready, notifying the team the video is available in the new language. Dubs where the tool flags low confidence on a segment, or where the source video has since changed, are routed to a person to check the translation before publishing. The agent never approves a dub for release on its own.