Edit videos with Claude and HyperFrames using plain editing terms

Jake Moran shows how to prompt Claude and HyperFrames for rough cuts, hooks, B-roll, a 3-screenshot launch video and style copying, using the right editing terms and as much context as possible.

Peter Yang · YouTube · published Oct 25, 2026 · Content & Creative

Edit videos with Claude and HyperFrames using plain editing terms — video by Peter Yang▶ Watch on YouTube
Prompt
Here is my raw recording. Do a rough cut: remove filler words and ums, cut pauses to a quarter of a second or less, drop repeated words and retakes, and use punch-ins to hide jump cuts. Add a zoom-in hook at the start and keep any music at 10% of my voice volume.

Visual breakdown

Jake Moran of HyperFrames shows that AI video editing works best when you use real editing terms and give the agent as much context as possible: a rough cut to remove filler, pauses and retakes, hooks with zooms and B-roll, and launch videos built from just three screenshots and a loose storyline.

The problem: Video editing is a wide field with its own jargon, and without the right words (and good timestamps) the AI cuts badly or wastes tokens.

From raw footage to a finished video

  1. Raw takes and referencesRaw recording, screenshots, blog links, brand kit, or a short reference clip to copy
  2. TranscriptionA local model like Parakeet gives accurate word timestamps for tight cuts
  3. Rough cutAgent removes filler words, long pauses, repeats and retakes
    Dropped: Ums, pauses, repeats, retakes
  4. Hook and effectsZoom-ins, B-roll, transitions like whip pans, music kept low
  5. Storyboard checkReview still frames before the full build, then leave notes on the video
  6. Finished videoLaunch video, talking-head YouTube video or short
InputsManual reviewAI agentResult

Editing terms worth knowing

  1. Rough cut: remove mistakes and get back to the storyline
  2. Jump cut and punch-in: cut repeats, then zoom to hide the jump
  3. Hook: zoom-ins, B-roll and speaker cards at the start
  4. Launch video in four parts: hook, context, payoff, brand

By the numbers

20:30 to 18:11raw vs edited length of one episode
98%Parakeet timestamp accuracy, per Jake
3screenshots to start a launch video
30-60 stypical launch video length

Try it yourself

Use HyperFrames (its desktop app, or its repo with any agent) and describe the edit in plain terms.

  1. Give the agent your raw footage plus as much context as you can: screenshots, a blog link, brand kit
  2. Ask for a rough cut: remove ums, cut pauses to a quarter second or less, drop repeats and retakes
  3. Ask for a hook: zoom-in, B-roll, speaker card, and name a few transitions, then let Claude pick
  4. Add music and say to keep it at about 10% of your A-roll volume
  5. Ask for a storyboard first and approve the still frames before the full build
  6. Leave notes on the video itself, then send the queued edits

Once a workflow works, ask the agent to save the steps as a skill so you don't repeat the work.

Tips from the conversation

  • Use terms both you and the model recognize, which gets results faster and saves tokens.
  • Models can't hear audio, so state music volume explicitly.
  • Bad timestamps cut words in half; a good local transcription model fixes that.
  • You still need your own idea and rough storyline; the AI fills in many choices itself.
  • For long videos, split into segments and stitch them to avoid reprocessing everything.

Distilled from the video’s captions

How to try it

  1. Open Claude.
  2. Copy the prompt and fill in the [placeholders].
  3. Watch Peter Yang's video for the full walkthrough and setup.

Related use cases

Share