Your video file
Use the unedited talking-head footage, without burnt-in captions.
The no-clue-where-to-start edition
AI video editing for absolute beginners.
The exact prompts to go from video file to first edit.
No editing language required. Copy a prompt, review a short preview, tell it what to fix, then save what it learns.
The first draft starts the process. Your corrections teach the style.
Before you type anything
You do not need an editing brief.
You need your footage, the point of the video and an AI coding agent that can work with local files.
One sentence about the point
Example: This video is about why AI should save us time, not create more homework.
Codex or Claude Code
Use a local coding agent that can read files, write the project and run the render. A normal text chat is not enough.
Setup / Do this once
Install two basics. Then HyperFrames.
HyperFrames needs Node.js 22 or newer and FFmpeg. The commands look technical, but you can copy them exactly.
Do not edit yet: Run the setup check and continue only when all six items say PASS.
Mac
Open Terminal. Check whether Homebrew is installed:
brew --version
If that fails, install Homebrew from its official instructions. Then run:
brew install node ffmpeg
Windows
Open PowerShell and run both commands:
winget install --id OpenJS.NodeJS.LTS -e
winget install --id Gyan.FFmpeg -e
If winget is missing, install or update App Installer from Microsoft.
Check the versions
node --version
ffmpeg -version
The Node number must begin with v22 or higher.
Create the HyperFrames project
In Terminal or PowerShell, move into the folder where you want the project, then run:
npx hyperframes init my-video
There is no separate global HyperFrames install. The npx command fetches and runs the project starter. Open the new my-video folder in Codex or Claude Code, then put a copy of your source video inside it.
Setup check / Run before editing
Before editing, audit this project. Confirm: 1. Node.js 22 or newer is available. 2. FFmpeg is available. 3. The HyperFrames CLI is available. 4. You can read [VIDEO FILE]. 5. You can write a small test file inside this project folder. 6. You can render a 3-second MP4 test without altering the source. Return PASS or BLOCKED for each check. If anything is blocked, stop and give me the exact fix. Do not begin the edit or switch to planning-only mode.
Step 1 / Kick off the edit
Start with the voice. Not the graphics.
Replace the two highlighted parts, attach or name the file, then send the prompt.
Your only decision: Does the shortened version still sound like the person speaking?
Prompt 1 / Make the first cut
I want you to edit [VIDEO FILE] as a vertical talking-head Reel. The point is: [ONE-SENTENCE POINT]. First, edit the voice for meaning and natural flow. Remove false starts, accidental gaps, repeated thoughts, detours and filler that makes the pace sag. Keep the speaker's natural rhythm, small imperfections, expressions, reactions, comic timing and any earned pauses. Never overwrite the original source file. Do not rearrange speech, remove a qualification or join words in a way that changes what the person meant. Flag any reordered sentence for approval. Do not add graphics yet. Give me the proposed cut and flag any join that may sound clunky or change the meaning.
Step 2 / Ask for a short preview
Do not build the whole thing yet.
A 15-second preview gives you something real to react to. A wrong direction is still quick to fix.
Your only decision: Does this feel like the right editing direction?
Prompt 2 / Build the first 15 seconds
Now create and render only the first 15 seconds as a 1080 by 1920 preview. Make the first 3 to 5 seconds the busiest part, then reduce the movement. Use quiet, readable captions inside the safe area. Add a larger editorial graphic only when it sharpens the point. Do not illustrate every sentence. Do not cover the speaker's eyes, mouth or important reactions. Clear a major graphic before the next idea arrives. Use one accent colour family. No random GIFs, constant zooming or decorative movement with no meaning.
Step 3 / Watch the preview
Watch it three times. One job each time.
Do not try to judge everything at once. Write down three things you want changed.
Ignore for now: Tiny timing, colour and spacing tweaks.
Listen without studying the graphics
Does the sentence make sense? Do any cuts sound abrupt or change the meaning?
Watch the captions and bigger visuals
Can you read them? Do they arrive with the speech? Does each larger visual earn its place?
Watch the person
Can you see the face, expressions and reactions when they matter? Is important text inside the safe area?
Step 4 / Give useful feedback
Tell it where. Tell it what.
Use timestamps. Say what should change and what should stay. Add the reason when it will help the next decision.
Example: 00:06 Keep the reaction. Slow the list reveal so each point has time to register.
Avoid: "Make it better." That gives the AI nothing reusable.
Prompt 3 / Turn your notes into an edit
Please revise the 15-second preview using these notes: [TIME] Change: [WHAT SHOULD CHANGE] Keep: [WHAT IS ALREADY WORKING] Reason: [WHY IT MATTERS] [TIME] Change: [WHAT SHOULD CHANGE] Keep: [WHAT IS ALREADY WORKING] Reason: [WHY IT MATTERS] [TIME] Change: [WHAT SHOULD CHANGE] Keep: [WHAT IS ALREADY WORKING] Reason: [WHY IT MATTERS] Render the revised 15-second preview before applying the direction to the full video.
Step 5 / Finish the edit
Approve the direction. Then finish it.
Once the revised preview feels right, apply that direction to the complete video.
Final check: Preview the export in Instagram before publishing because the interface can vary.
Prompt 4 / Apply it to the full video
The revised preview is the approved direction. Apply the same editing decisions to the full video. Keep captions quiet and readable. Use roughly 6 to 9 designed events across a 50 to 60 second Reel, adjusted for the actual length. Reveal list items when they are spoken and leave them up long enough to register. Use screenshots, interface cards, cutaways or reactions only when they prove the claim, make an example concrete, sharpen a joke, hide a difficult dialogue edit or change the visual plane. Return to the speaker after cutaways. Keep essential text inside the safe area. For a 1080 by 1920 Reel, keep 90 px clear on the left, 180 px on the right, 250 px at the top and 420 px at the bottom. Render the full draft for review.
Step 6 / Make the next edit easier
Save what the AI learned from you.
This is the part that stops every new edit beginning from nothing.
The boundary: A one-off choice should not become a permanent rule unless you approve it.
Prompt 5 / Turn corrections into rules
Review my approved changes from this edit. List only the decisions that could usefully apply to a future talking-head video. Write each one as a clear editing rule. Separate: 1. Rules that should apply every time 2. Preferences to try again but review 3. Choices that belonged only to this video Add the approved reusable rules to my saved video editing standard. Do not overwrite existing rules or turn one-off choices into permanent rules without asking me.
Step 7 / Check the actual export
Watch the export. Not just the timeline.
A successful render is not the same as a checked video. This prompt separates production errors from choices that still need you.
Nothing publishes here: The agent may fix objective errors, but it must not publish or share the video.
Prompt 6 / QA the exported MP4
Review the actual exported MP4 against the original source and approved preview. Watch it from beginning to end with sound on. Then inspect key frames with sound off. Check: - spoken meaning still matches the source - audio and picture remain in sync - cuts sound natural - captions arrive with the correct words - text is readable on a phone and inside the safe area - face, expressions and gestures stay clear - larger graphics have useful jobs - the file opens and plays correctly Return PASS items, FIX items with exact timestamps, and subjective choices needing my decision. Fix objective production errors. Do not publish or share anything.
The whole beginner loop
Your first draft starts the learning.
Install and test the tools. Name the file and point. Edit the voice. Render 15 seconds. Give three timestamped notes. Finish the video. Save the reusable rules. QA the actual export.
Download the printable PDF