Make a rough cut from a long training video

They used Claude Code to transcribe a 6-hour session locally, remove dead air and mistakes, and prepare a Kdenlive project for cleanup.

Done with Claude Code· Intermediate· Nov 2026

What they did
Feng Yu loaded two Claude Code skills, videocut-skills and cli-anything-kdenlive, and asked the agent to merge them without relying on Volcano Engine’s paid transcription service. The agent built a local process that used faster-whisper for speech-to-text, then applied the editing rules to mark pauses, slips of the tongue, and other sections to drop. It also wrote out a Kdenlive project file so Feng could open it and finish the edit by hand.
How it went
The recording was transcribed and turned into a usable rough assembly, which reduced the work from around 3 days to just over 1. The draft still left in a lot of repeats and verbal mistakes, and the project contained too many individual clips. Because the computer could not run faster-whisper large-v3, Feng used the small int8 model in parallel chunks, which made the cut points less exact.
Tips
This is a good fit when you want an AI-assisted first pass rather than a final edit. If your hardware is limited, expect some loss of precision and plan extra time for manual tightening.

Try it yourself with Claude Code

I have a [length]-long recording saved at [file path]. Stay local if you can: transcribe it, flag pauses, filler speech, off-topic stretches, and [other items to remove], then build a [Kdenlive or similar editor] project that leaves those parts out for later review. Keep the source file unchanged. Before any saving, replacing, submitting, deleting, or exporting, show me the cut plan and wait for my approval. When you are finished, give me the project file plus a timestamped list of the edits.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.