Transcript-Driven Video Edit Pipeline (HyperFrames)

Your automated edit "works", but the result still has problems. Captions drift out of sync after minute three. Every cut makes a tiny click. The lower third shows up two seconds late. And changing one title means waiting through a full re-render. This skill makes Claude Code run the edit the way a technical editor would. It builds one edit decision list that every later step follows, and it remaps every timestamp to the new timeline. It re-renders only the parts you changed.

What it does

• Builds a rough cut from word-level timestamps plus silence detection. It cuts dead air and repeated takes, and catches the "um"s that transcripts tend to leave out.
• Keeps a single JSON edit decision list as the source of truth, so captions, graphics, and re-renders all stay in sync
• Remaps caption and graphic timing from the raw footage to the cut timeline, with a correction dictionary for brand names and jargon
• Places branded graphics (lower thirds, titles, callouts) on the exact words they belong to, using brand preset tokens and HTML compositions
• Splits the timeline into cached segments so a revision re-renders only what changed. Audio is rendered as one continuous pass, so the joins don't gap.
• Produces a cut report you can review as text: every removal, its reason, and the words removed

What you get

A working pipeline in your project: normalize, transcribe, edit list, rough cut, graphics, captions, loudness-matched export. You also get the exact commands and a cut report you can check before you render the final.

Best for

Creators and small teams who edit talking-head, interview, tutorial, or podcast-video footage with Claude Code and want repeatable edits on brand without scrubbing a timeline.