Skip to main content
aifeatures.ai
Teams shipping talking-head explainers

The studio avatar pipeline

Claude writes, ElevenLabs is the voice library, HeyGen lip-syncs the consented face, Descript masters.

Who owns what

  1. Claude

    The script — 60–90 seconds, pronunciation notes. Not the face and not the WAV.

    Try Claude (opens in a new tab)

  2. ElevenLabs

    The voice id you reuse outside HeyGen. HeyGen owns picture, not a second clone.

    Try ElevenLabs (opens in a new tab)

  3. HeyGen

    The digital human — lip-sync, language take. Not the script and not the timeline.

    Try HeyGen (opens in a new tab)

  4. Descript

    The master — captions, B-roll overlay, export. Not a second avatar generator.

    Try Descript (opens in a new tab)

How to use this stack

  1. Claude writes into a sheet. Status stays draft until a human marks approved.
  2. ElevenLabs renders the VO for that language. One voice id.
  3. HeyGen animates the approved avatar against that track. No celebrity faces.
  4. Descript captions and exports. Publish is a separate step until the avatar is trusted.

Why we picked this: HeyGen + n8n: script to talking avatar without living in the GUI