clipbrief-weld.vercel.app

transcribe_video

Return the transcript for one clip. Clips that ship a caption track are parsed from it; the rest were transcribed with Whisper running locally at build time, and those responses carry live:false plus the model used. Results are committed, so this is instant and costs nothing. If you only need a transcript that already exists, get_video is equivalent and clearer.

Page: /

Input schema

{
  "type": "object",
  "properties": {
    "video_id": {
      "type": "string"
    },
    "force": {
      "type": "boolean",
      "description": "Ignored for the speech-to-text path, which is build-time only."
    }
  },
  "required": [
    "video_id"
  ]
}