Video Editing

Fast cut

Quick cut of ums and silences into a 9:16 short

Picture shape: 9:16
Runs on Claude CodeCodexAntigravity

Hand your agent one recording of one person talking and get a clean 9:16 short with the ums, silences and restarts cut out, karaoke captions and a hook title.

Important: this is a skill for terminal AI coding agents (Claude Code, Codex) that runs on your own computer. It is not a prompt for the ChatGPT website. Check the requirements below before buying.

Why this look

You recorded a talking-head clip, a vlog or a podcast. Cutting every pause and "à, ờ, ừm" by hand takes longer than the recording. This look does the cut with scripts and ffmpeg, and you decide the story and the crop.

  • Removes silences, Vietnamese fillers, stutters and restarted phrases at a level you choose: light, medium (default) or strong.
  • Cuts land on whole frames on a 30 fps grid, so a phone clip with a variable frame rate does not drift.
  • Real footage only: no AI video, no generated faces. Nothing appears on screen that the speaker did not say.

Three outputs from one recording

  • One short (9:16): the agent reads the transcript and picks one self-contained story window, hook line first. Default length 45-75 seconds.
  • Several shorts: one window per short from a long talk, transcribed once.
  • Long clean version: the whole recording, original frame kept, with an SRT file.

What a short can carry

  • Karaoke captions placed above the bottom of the frame, never on the face.
  • A hook title on every short: static, a drop with a small shake, or a shatter into shards.
  • Optional highlight cards on spoken words (keyword, stat, list, tag), photo b-roll with a face bubble, and a bundled music bed.
  • An optional effects library of 133 effects (text behind the head, rolling numbers, stamps, comment box and more), picked per spoken line.

How it works

Open your agent in the folder with your recording and type:

Cắt clip này thành 1 short 60 giây, mức cắt medium, có phụ đề karaoke

The agent transcribes the recording, proposes the cuts, reads what was removed and fixes any word that got chopped. It checks stills to place the crop, renders, then runs a check on frame count, audio and video length, duplicate frames and loudness. It reports what was cut, in seconds.

What you need

  • Claude Code or Codex.
  • Python 3.10 or newer and ffmpeg. An NVIDIA GPU encoder (h264_nvenc) is optional; without it the skill uses libx264, slower.
  • OpenCV in a small Python environment, used only to find the speaker's head.
  • Edge or Chrome, used only to draw the cards.
  • SONIOX_API_KEY (or ElevenLabs Scribe) for speech-to-text.

Install

Copy the skill folder into your skills folder:

~/.claude/skills/talkinghead-fast-cut/

For Codex, use ~/.agents/skills/talkinghead-fast-cut/.

Who it is for

A fast cut of one talking-head, vlog or podcast recording: one short, several shorts or a long clean version.

Not a fit if you want

  • A faceless video.
  • Designed motion graphics or overlay editing.

Frequently asked questions

Can I add my own logo at the end?

Yes. An optional end card of about 3 seconds holds the last frame and shows your message and logo. It is off unless you turn it on.

Can I use the videos commercially?

Videos you render are yours for any commercial use. You may not resell, redistribute or share the skill files.

More

Related looks