edit-dialogue skill
Install this skill with npx skills add synchronicity-labs/skills --skill edit-dialogue, or get every skill with the Claude Code plugin. See Agent skills.
What it does
edit-dialogue fixes a few spoken words in a video without a reshoot. It transcribes the video, maps your fix to the exact words, synthesizes the new words in the speaker’s own voice, lets you hear a preview, then re-syncs the lips on just that part. Everything else keeps the original picture and sound.
When to reach for it
Try
- “In ./promo.mp4 I say 49 now, fix just that.”
- “Cut the ‘um’ after ‘billed yearly’ in this video”
- “The name should be Siobhan, not Shannon”
What you need
- The sync. MCP server connected. The setup skill walks you through it. The REST version of this workflow is in early access; see Transcription and Dialogue Editing.
- A video with one speaker, up to 10 minutes.
Common questions
Why does it show me the transcript?
So you can both point at the same words. Every change and removal is tied to a word in the transcript, which keeps the edit exact.
Does the preview cost money?
It can. The preview clones the speaker’s voice and synthesizes the new words, so the agent mentions it when confirming the edits. The final video is quoted separately before it’s created.
A name came out wrong.
Ask for a pronunciation. The agent adds a respelling like “shivawn”, written as a plain word, and makes a new preview.
Some edits failed.
Usually the new wording is too long for the gap it fills. The agent says which edits failed and suggests shorter wording.
Can I replace a whole paragraph?
This workflow is for targeted words. For a new script, lipsync with a voice is the better tool, and the agent will say so.
It’s working if
- The preview audio has only your words changed.
- You approve the preview and the cost before the video is made.
- The output plays the original video with just those words different.