Edit Videos with Claude Code

Turn a raw talking-head clip into a creator-style short with Claude Code.

Introduction

30 Second Summary

A creator’s videos can feel fast, clear, and polished within seconds. Reproducing that feeling in your own footage usually means guessing where every cut, caption, zoom, and visual should go.

In this project, you will use Claude Code to measure how a creator edits their videos. You will turn your raw talking-head footage into a finished short-form edit with clean cuts, timed captions, supporting visuals, sound cues, and a reusable personal taste rule.

Claude Code turns source clips and an editing prompt into a captioned video timeline you can review and render.

What You'll Build

Your finished video plays with the creator’s measured rhythm while keeping your face visible, your captions readable, and your own editing preferences in control.

By the end of this project, you'll have:

A finished creator-style video that you can play, review, and share from edits/<name>/render.mp4.
A measured creator profile built from five hand-picked videos that share the style you want to recreate.
A reusable taste rule that changes future edits according to your own creative judgement.
Secret Mission: Render your edit on GitHub Actions for free: a private repo, a render in the cloud, and the finished video downloaded back.

Are there any prerequisites?

You need Claude Code with a Claude subscription on macOS, Windows, or Linux and about 2 GB of free disk space. In Step 3, you can record a 30-second to 3-minute talking-head video, use an existing MP4 or MOV file, or download the demo video. A GitHub account is only required for the optional Secret Mission.

Before We Start

Before We Start...

Before the hands-on work begins, lock in the result you are working toward. You will study a creator’s editing style, apply it to a talking-head video, and refine the result with your own creative preference.

Install and Set Up the Editor

Your finished video needs a workspace that keeps every creator profile, cut, and render together. This setup gives the editor one predictable place to work.

The editor cannot measure your creator until its plugins are installed. Claude Code also needs a checked toolchain before it can cut or render your video.

In this step, get ready to:

Prepare a new empty Claude Code project folder.
Install the NextWork editor plugins.
Complete the guided setup until the editor doctor reports Ready.

Prepare Claude Code for the editor

A clean project folder gives Claude Code one place for every output created later. The NextWork plugin marketplace provides the editor tools for that workspace.

Open a terminal first. You will use it to check Claude Code and start the editor from the correct folder.

macOS

Press Cmd+Space to open Spotlight.
Type Terminal into the search bar.
Press Return to open Terminal.

Windows

Press the Windows key to open Search.
Type PowerShell into the search bar.
Press Enter to open PowerShell.

Linux

Press the Windows key to open your application search.
Type Terminal into the search bar.
Press Enter to open Terminal.

Check whether Claude Code is installed by running this command:

claude --version

Choose the tab that matches what your terminal shows.

✔️ I see a version number

Claude Code is installed and ready. Continue below to create its working folder.

ⓧ Command not found

Claude Code needs to be installed before it can open your workspace. Choose your operating system for Anthropic's current native installer.

macOS

Install Claude Code with Anthropic's native installer by running this command:

curl -fsSL https://claude.ai/install.sh | bash

Confirm the installation by running this command:

claude --version

You should see a Claude Code version number.

Windows

Install Claude Code from PowerShell by running this command:

irm https://claude.ai/install.ps1 | iex

Confirm the installation by running this command:

claude --version

You should see a Claude Code version number.

Linux

Install Claude Code with Anthropic's native installer by running this command:

curl -fsSL https://claude.ai/install.sh | bash

Confirm the installation by running this command:

claude --version

You should see a Claude Code version number.

Claude command still missing?

Close your terminal after the installer finishes.
Open a new terminal using the platform instructions above.
Check Anthropic's installation troubleshooting guide for path and installer fixes.
Ask for help using Help me fix the missing Claude Code command.

Move to your Desktop
Create the ai-video-editor folder and move into it by running these commands:

cd ~/Desktop
mkdir ai-video-editor
cd ai-video-editor

Confirm the new folder contains no project files by running this command:

ls

The command should return no project files. Your clean folder now gives every project output one predictable home.

The first Claude Code launch may ask you to sign in through your browser before the session opens.

Start Claude Code inside the empty ai-video-editor folder by running this command:

claude

Your clean workspace is ready. Claude Code now uses the empty ai-video-editor folder as its working folder.

Add the NextWork plugin marketplace by typing this slash command into Claude Code:

/plugin marketplace add nextwork-projects/ai-video-editor

What does this command do?

This command registers the NextWork marketplace with Claude Code. That marketplace makes the editor plugin available for installation.

Confirm Claude reports that the NextWork marketplace is available.

Marketplace not being added?

Check that you entered the command in Claude Code instead of a separate terminal.
Check that the repository name matches the command exactly.
Ask for help with the marketplace command using Help me troubleshoot why Claude Code cannot add the NextWork plugin marketplace.

Install the editor plugin by typing this slash command into Claude Code:

/plugin install ai-editor@nextwork

What does this command do?

This command installs the editor from the NextWork marketplace. The installation also makes the bundled creator teardown plugin available.

Select Install for you, in this repo only (local scope) when Claude Code asks where to install the plugin.
Press Enter to confirm the local scope.

Confirm Claude reports that the plugin installation completed.

Check for the bundled skill by typing part of its command into Claude Code:

/creator

You do not need to press Enter. If Claude Code shows /creator-teardown:creator-teardown in the suggestions, press tab to select that option.

Plugin installation not completing?

Confirm the marketplace command completed before retrying the plugin command.
Check that the plugin name ends with @nextwork.
Ask for installation help using Help me troubleshoot why ai-editor@nextwork is not installing in Claude Code.

Install the checked toolchain

The setup skill checks every dependency that the editor needs. It proposes an installation when a required tool is missing or outdated.

What does the toolchain include?

Python 3.9 or newer runs the editor scripts.
Node.js 20 or newer runs Remotion 4.0.530.
The ffmpeg tool handles video processing.
Git supports the editor's project dependencies.
faster-whisper uses the small model for free local transcription.
The required Python packages live in the editor's local virtual environment.

Start the guided dependency setup by sending this normal message to Claude Code:

set up the editor

What happens during setup?

Claude checks each required tool before proposing an installation.
Claude asks for permission before each installation.
The setup creates ~/.ai-video-editor/ as the shared local tool home.
The tool home stores the virtual environment in venv/.
The tool home stores the renderer in remotion/.
The tool home stores the transcription model in models/.

The first full setup downloads about 1.5 GB. A quiet pause is normal while the renderer downloads.

The setup skill adjusts its commands for your operating system.

macOS

The setup skill uses python3 on macOS.

Approve each required installation when Claude asks for permission.
Run any Next steps lines that Homebrew prints.
Return to the same Claude Code session after those lines finish.

Windows

The setup skill uses py on Windows.

Approve each required installation when Claude asks for permission.
Close Claude Code after any winget installation finishes.
Start Claude Code again with the same empty project folder.
Repeat the setup message from above.

Linux

The setup skill uses python3 on Linux.

Approve each required installation when Claude asks for permission.
Enter your computer password when a command beginning with sudo requests it.
Let each installation finish before responding to the next prompt.

The setup result falls into one of these three states.

✔️ I see supported versions

Good progress. Python 3.9 or newer is ready.

Node.js 20 or newer is also ready. Continue to the final doctor check.

Ask Claude Code to check both setup components by sending this normal message:

check that the Python packages and transcription model are ready

Confirm the python packages result starts with ok and points to ~/.ai-video-editor/venv.
Confirm the transcription model result starts with ok and shows faster-whisper small.

Confirm the setup reports that the Remotion renderer is available.

ⓧ I see an older version

An older Python or Node.js version cannot satisfy the editor's checked requirements. The setup skill provides the supported upgrade for your computer.

Approve the upgrade proposed by the setup skill.
Complete any operating system restart instruction shown above.
Return to Claude Code in the same project folder.
Repeat the setup message from above.
Confirm Python reports 3.9 or newer.
Confirm Node.js reports 20 or newer.

ⓧ A required command is missing

A missing command means the corresponding tool is not available on your computer yet. The setup skill identifies that tool before requesting installation permission.

Approve the installation proposed by the setup skill.
Complete any operating system follow-up instruction shown above.
Return to Claude Code in the same project folder.
Repeat the setup message from above.
Confirm the previously missing tool now passes its check.

Finish with a clean doctor result

The editor doctor checks the finished environment as one system. Before you run it, do you expect every required dependency to pass now?

Run the final editor check by sending this message to Claude Code:

run the editor doctor

What does the doctor check?

The doctor checks Python against the editor's minimum version.
The doctor checks Node.js against the renderer's minimum version.
The doctor checks ffmpeg.
The doctor checks Git.
The doctor checks the required Python packages.
The doctor checks the local transcription model.
The doctor checks the Remotion renderer.

When setup is complete, every required line reads ok. The final line reads Ready.

Seeing a FIX line?

Run the editor doctor again.
Stop only when the final line reads Ready.
Ask for help with the failed check using Help me resolve the FIX line from the AI Video Editor doctor.

That setup hurdle is cleared. Your editor now has every checked dependency it needs to measure creators or process video.

✔️ Awesome, I've got everything!

Your editor setup is complete. Keep Claude Code working in the same empty project folder.

ⓧ I'd like to double check the full code

This step does not add application code. Double-check these exact setup artifacts instead:

Claude Code plugins: ai-editor@nextwork is installed. The bundled creator-teardown plugin is available.

Claude Code project folder: A new empty folder has been created. That folder is open in Claude Code.
Project outputs: No creator teardown or edit output has been created yet.

Your editor is ready for real footage. Next up, you will measure the creator habits that shape your edit.

Break Down a Creator

Your editor is ready. Now you need a reference style that gives it concrete editing targets.

One favorite creator can still use several formats. In this step, you will choose five videos that share the same format, then let Claude Code measure the patterns they repeat.

In this step, get ready to:

Choose five videos from one favorite creator that use a similar format.
Paste the five video links into Claude Code.
Review the shared pacing, caption, layout, hook, and visual patterns.

Choose five similar creator videos

The teardown accepts public TikTok or YouTube Shorts links. Instagram is login-walled for this workflow, so use the creator's TikTok or YouTube version when possible.

Your local faster-whisper small model transcribes the selected videos. This keeps transcription on your computer.

Open your favorite creator's profile on TikTok or YouTube.
Choose five public short-form videos that use the same format.
Copy the URL of each chosen video.

Build the creator breakdown request with all five links by using this message:

Break down these five videos because they use a similar editing style I want to learn from:

[[CREATOR_VIDEO_1_URL="first public TikTok or YouTube Shorts URL"]]
[[CREATOR_VIDEO_2_URL="second public TikTok or YouTube Shorts URL"]]
[[CREATOR_VIDEO_3_URL="third public TikTok or YouTube Shorts URL"]]
[[CREATOR_VIDEO_4_URL="fourth public TikTok or YouTube Shorts URL"]]
[[CREATOR_VIDEO_5_URL="fifth public TikTok or YouTube Shorts URL"]]

What does this request do?

The five links tell the teardown exactly which videos to download and analyze.
The opening sentence tells Claude to treat them as one style sample.
A focused sample makes repeated editing choices easier to spot.

Paste the completed message into Claude Code.
Press Enter to start the teardown.

This can take a few minutes because Claude downloads and measures every clip. Quiet terminal output during that wait does not mean it has stalled.

You'll see Claude start processing the five links you selected.

Teardown not starting?

Open each copied URL in your browser to confirm that the video is still public.
Replace any unavailable link with another video that uses the same format.
Help me find the link blocking my creator teardown.

Confirm the creator profile files

Press Cmd+Space (macOS) or the Windows key (Windows) to open system search.
Type Finder (macOS) or File Explorer (Windows).
Press Enter to open the app.
Open the Desktop/ai-video-editor/creator-teardowns folder.
Open the folder named after your creator.
Confirm that the folder contains teardown.md and style.json.

Leave the creator folder open so you can inspect the files after the consistency check.
Return to Claude Code.

What are these files for?

teardown.md explains the hooks, pacing, structure, captions, and visual patterns shared by the five selected videos.
style.json stores the measured pace, zoom, caption, sound, and layout settings Claude applies during the visual style pass.

Check that all five videos belong to one consistent style by sending this message to Claude Code:

Review teardown.md against the five source videos. Confirm whether all five videos share a consistent editing style across hooks, pacing, structure, captions, and visuals.

For each category, list the matching choices and the source video URLs that demonstrate them. If any video differs from the rest, label that video MISMATCH and explain which choices do not fit.

You’ll see matching source URLs under each style category. If one video does not fit the group, Claude Code labels it MISMATCH and explains the differences.

✔️ All five videos match

Claude Code confirms that all five videos share one consistent editing style. Continue below to open the finished style profile.

ⓧ Claude labels a video MISMATCH

Identify each video labelled MISMATCH in Claude Code's response.
Choose another public video that matches the other four videos.
Replace the mismatched video's URL in your five-link teardown request.
Send the updated request to Claude Code.
Repeat these actions until Claude Code confirms that all five videos share one editing style.

Return to the creator folder you left open.
Open teardown.md.
Read the hook, pacing, structure, captions, and visuals sections.
Compare the reported patterns with the five source videos.
Open style.json.
Review the pace, zoom, and captions settings that Claude applies later.

Your creator profile is ready. Claude uses it during the style pass to shape your edit.

Handoff files missing?

Scroll back to the end of the teardown output in Claude Code.
Look for DONE: 5 videos analyzed.
Send the complete five-link teardown message again if that line is missing.
Help me find the creator teardown files.

Your creator profile now gives the editor measured targets for visual rhythm. Next, you'll clean the spoken story in your unchanged original recording.

Cut Your Video

Your creator teardown gives the editor measured targets. A raw talking-head recording with natural pauses and repeated attempts gives those targets footage to shape.

In this step, Claude Code uses a last-take-wins edit to keep each clean final attempt. You will judge the result by watching the rendered first cut before approving it.

In this step, get ready to:

Choose a talking-head MP4 or MOV file from one of three sources.
Watch the complete first cut from beginning to end.
Approve the cut or send all timestamped corrections in one message.

Choose and send Claude your video

Record a new video

Record yourself speaking about a familiar topic for 30 seconds to 3 minutes.
Pause for a few seconds during the recording.
Restart one sentence from the beginning after a deliberate false start.
Save the recording as an .mp4 or .mov file inside your ai-video-editor folder.

Use one of my videos

Choose an existing .mp4 or .mov video of yourself speaking for 30 seconds to 3 minutes.
Check that the video includes a natural pause, filler, or restarted sentence.
Copy the selected video into your ai-video-editor folder using Finder (macOS) or File Explorer (Windows).

Use the demo video

The official GitHub release includes a raw sample with pauses and retakes for this workflow.

Download sample-take.mp4 from the official repository release.
Move sample-take.mp4 into your ai-video-editor folder using Finder (macOS) or File Explorer (Windows).

You should now have an .mp4 or .mov talking-head video inside ai-video-editor. When you play it, you should hear clear speech with at least one pause, filler, or restarted line for the cut workflow to remove.

Copy the video's full path by following the instructions for your operating system:

macOS

In Finder, open Desktop/ai-video-editor.
Right-click your video file.
Press and hold Option.
Select the menu item that ends with as Pathname.

Windows

In File Explorer, open Desktop/ai-video-editor.
Press and hold Shift.
Right-click your video file.
Select Copy as path.

Paste the copied path here: full path to your raw MP4 or MOV video.

The same edit request works on macOS, Windows, and Linux. Only the saved file path changes between operating systems.

Start the cut workflow with your saved path by sending this message in Claude Code:

edit my video: [[RAW_VIDEO_PATH="full path to your raw MP4 or MOV video"]]

What does this request do?

The edit my video: phrase starts the cut workflow.
Your full path identifies the video Claude needs to transcribe.
The workflow leaves your original video unchanged.

Send the completed message in Claude Code.

You'll see Claude begin transcribing the source video.

Claude cannot find your video?

Check that the path includes every folder leading to the video file.
Check that the filename ends in .mp4 or .mov.
Remove any accidental spaces from the beginning or end of the path.
Help me check the full path to my video.

Claude transcribes the spoken words after accepting the path. It creates the generated edit in a new folder inside edits/.

Review the first cut

Claude uses your transcript to create the first cut. Once the render is ready, you can review the timing and decide what still needs to change.

Watch the whole video before deciding whether the first cut is ready. Listening from beginning to end makes awkward transitions easier to catch.

Wait for Claude to finish transcribing the source video.
Wait for Claude to render the first cut.
Play the rendered video from beginning to end.

You'll hear a tighter spoken story with false starts or dead pauses removed. The final intended takes remain in place.

You'll also see an early visual pass that borrows some choices from the creator profile. Claude has not yet been explicitly told to follow every measured rule in style.json.

Does a cut sound wrong?

Replay the section before the awkward transition.
Note the approximate timestamp where the problem begins.
Describe the words or pause that should be removed, restored, or shortened.
Help me write a precise timestamped correction.

Approve or correct the first cut

Your approval locks the spoken story before styling begins. If anything sounds wrong, collect every correction first so Claude can rebuild the cut once.

✔️ The cut sounds right

Approve the clean cut by sending this message in Claude Code:

The first cut sounds right. Approve this cut.

Claude should confirm that the spoken edit is approved and ready for styling.

ⓧ I found changes

Watch the entire first cut before replying. A single batch lets Claude apply every correction during one rebuild.

Note the approximate timestamp where each issue begins.
Describe one clear change for each timestamp.
Combine every correction into one message using this pattern:

Please update the first cut with these changes:
- Around 0:04, remove the false start where I begin the sentence twice.
- Around 0:18, shorten the pause before the next sentence.
- Around 0:31, restore the final words of the sentence.

Replace the example timestamps and instructions with the issues you found.
Send the completed message to Claude Code once.
Wait for Claude to rebuild the cut.
Watch the corrected video from beginning to end.

You should hear every requested correction in the rebuilt video. The intended final takes should remain intact.

Was a correction applied to the wrong moment?

Replay the problem section and note a more precise timestamp.
Quote a few nearby words so Claude can identify the correct take.
Help me refine one timestamped correction.

Approve the corrected cut by sending this message in Claude Code:

The corrected cut sounds right. Approve this cut.

Claude should confirm that the corrected spoken edit is approved and ready for styling.

The playback contains your approved spoken story with some creator-inspired styling. The full set of measured rules remains for the next style pass.

Confirm that Claude Code reports the spoken cut as approved.
Play the approved cut from the beginning.
Pause on a frame that clearly shows the video player and timeline.
Capture a screenshot of the approved cut in playback.
Save the screenshot with your project evidence.

You have cleared the hardest editorial decision. Your spoken story is now locked before Claude follows the full creator-style rule set.

Your spoken story is clean and approved. Next, you'll explicitly tell Claude to follow the measured caption, pacing, zoom, visual, sound, and layout rules from the creator profile.

Style, Review, and Render on Your Laptop

Apply the Creator Style and Render

Your approved cut now tells the spoken story cleanly. Step 4 applies the measured rules from your chosen creator to this video.

In this step, Claude Code reads the saved profile for @creator handle without @. The profile supplies the caption, zoom, visual, sound, and placement rules for this video.

This step uses someone else's measured style to finish the current video. Step 5 is where you save one of your own choices as a reusable rule.

In this step, get ready to:

Apply the selected creator's measured rules to this video.
Review the generated stills before rendering.
Tell Claude to render the video after the stills match your intent.

Apply the creator's measured style

The measured profile turns your creator research into editing decisions. This style pass owns the captions, zooms, supporting visuals, sound cues, and on-screen placement around your face.

Confirm the creator handle you saved in Step 2: creator handle without @.
Apply the measured profile by sending the personalized message below in Claude Code:

Edit it in the style we analyzed for [[CREATOR_HANDLE="creator handle without @"]]. Open the generated stills for me to review. Wait for my approval before rendering the video.

What does this message do?

The saved creator handle identifies the measured style profile.
Claude uses the approved cut and its word timings as the starting point.
Claude applies the measured caption, zoom, and sound patterns.
Claude chooses supporting visuals that match the spoken moments.
Claude opens the generated stills and waits for your approval before rendering the video.

Review each generated still for caption placement, visual cards, and face clearance.

Describe any changes you want before rendering.
When the stills match your intent, tell Claude to render the video by sending this message:

The stills look right. Render the video and then open it up.

Review the finished edit

Choose the tab that matches what happened after Claude finished editing.

✔️ The video opened automatically

Watch the finished video from start to finish.
Check that supporting visuals match the spoken moments.
Confirm that captions leave your face visible.
Confirm that visual cards leave your face visible.

Identify one creative choice you want to personalize in the next step.

You should see a complete creator-style video while your approved spoken story remains intact.

ⓧ The video did not open

Press Cmd+Space (macOS) or the Windows key (Windows) to open your search bar.
Type Finder (macOS) or File Explorer (Windows).
Open the Desktop folder.
Open the ai-video-editor folder.
Open the edits folder.

Open the newest folder for your current edit.
Double-click render.mp4 to play the finished video.

Your finished video lives inside Desktop/ai-video-editor/edits even when it does not open automatically.

Can't find the finished video?

Sort the folders inside edits by the most recently modified date.
Open the newest folder to look for render.mp4.
Help me find the finished video.

You have finished one video by following the creator's measured rules. Next, you'll save one of your own preferences so it can override conflicting creator rules in this video and future edits.

Teach the Editor Your Taste

Save Your Own Editing Preferences

Step 4 applied the creator's measured rules to one finished video. Now you will use that result to decide where your own taste should take priority.

In this step, Claude Code saves one genuine correction as a reusable preference. That preference overrides any conflicting creator-profile rule in this video and future edits.

In this step, get ready to:

Choose one place where your taste differs from the creator profile.
Save your correction as a reusable editing preference.
Re-render the current video to confirm that your preference takes priority.

Save a preference that overrides the creator profile

Examples of useful taste rules

A useful taste rule names one visible choice and the direction you want it to change.

Make the captions larger and highlight the word being spoken.
Add slow zooms during long talking sections.
Keep visual cards above my shoulders so they do not cover my face.
Use sound effects only when a new section begins.
Replace the polished animations with hand-drawn doodles using loose ink outlines, simple shapes, paper texture, and one accent colour.

For visual references, explore Adobe's animation drawing techniques and hand-drawn comic animation example.

Choose one change you genuinely want after watching your finished video.
Enter your correction here: describe the change you want.
Copy the personalized correction below:

[[EDIT_CORRECTION="describe the change you want"]]. Save this as a reusable taste preference and apply it to the current edit. Render the updated video and open it for me to review.

What does this message do?

The message identifies one concrete visual or audio choice you want to change.
The taste skill converts your feedback into a reusable editing rule.
Your saved rule takes priority when a future creator profile makes a conflicting choice.
Claude applies the rule, renders the updated video, and opens it for review.

Send the personalized correction to Claude Code.

Claude should show the plain-language rule it derived. The rule should describe a repeatable behavior for future edits.

Does the preference feel too broad?

Name the exact visual or audio element you want to change.
Describe the direction you want that element to move.
Help me write a specific taste rule.

Review the corrected render

Claude saves your rule for future edits, then rebuilds the current video. The corrected render proves that your preference overrides the creator profile.

The corrected laptop render can take about as long as the render you completed earlier. A quiet terminal during frame generation is expected.

Wait for Claude to finish the local re-render.
Watch the video Claude opens from start to finish.
Check the part of the video where your correction should appear.

You should see your correction in the finished edit. The parts you did not ask to change should still match the approved version.

Correction Missing From the Render?

Tell Claude the exact timestamp where the old behavior remains.
Ask Claude to rebuild the affected plan with the saved taste rule.
Help me check why my saved taste preference is missing from the corrected render.

That is a real editing win. Your personal preference now changes the video itself.

Your editor now applies a preference that came from your own creative judgement. Future edits can start from that same direction.

Secret mission

Render on GitHub Actions for Free

Move long video renders off your laptop with GitHub Actions. A private repository holds the files while a cloud machine processes the edit and sends the finished video back to your project.

Clean Up Your Resources

The main project runs locally, so it has no ongoing cloud cost. If you completed the Secret Mission, your private GitHub repository remains stored until you delete it, while the downloadable render artifact expires after seven days.

Resources you used:

The Claude Code project folder at ~/Desktop/ai-video-editor, including creator profiles, approved cuts, finished videos, edits/<name>/github-render/, and edits/<name>/github-render-media.zip.
The editor tool home at ~/.ai-video-editor, including the local environment, renderer, models, and saved taste.
The screenshots of your approved first cut, finished styled video, corrected render, and successful cloud render.
The private GitHub repository, the render workflow, and its render artifact, if you completed the Secret Mission.

Keep everything running

No action is needed. Choose this if you are still editing videos or want to run the private workflow again.

Keep ~/Desktop/ai-video-editor so your creator profiles, edits, and finished videos remain available.
Keep ~/.ai-video-editor so the editor can reuse its renderer, models, and saved taste.
Keep edits/<name>/github-render/ inside the project folder. There is no separate github-render folder on your Desktop.
Keep the Secret Mission repository private while it contains your footage.
Download the render artifact before its seven-day retention period ends if you want another copy.

Pause - I'll come back to this later

Shut down local processes while keeping the project ready for later. The GitHub workflow only runs when the editor triggers it.

Close Claude Code.
Keep ~/Desktop/ai-video-editor.
Keep ~/.ai-video-editor.
Keep the Secret Mission repository private.
Open the private repository's Actions tab if you want to delete the cloud copy before it expires.
Select the render workflow.
Open the completed render run.
Click the delete icon beside the render artifact under Artifacts.

The artifact is gone when render no longer appears in the run summary. Your local copy at edits/<name>/render-github.mp4 remains available.

Delete - I don't want to use this again

Deleting the private repository and local folders removes the remaining copies created for this project. Save any finished video you want to keep somewhere else before continuing.

Open the Secret Mission repository on GitHub if you created it.
Click Settings under the repository name.
Scroll to Danger Zone on the General settings page.
Click Delete this repository.
Click I want to delete this repository.
Click I have read and understand these effects.
Type the repository name into the confirmation field.
Click Delete this repository to remove the repository, release file, workflow runs, and artifact.

That removes the private cloud copy. The next commands permanently remove the local project and editor state from your computer.

macOS

Close Claude Code.
Remove the project folder and editor tool home by running this command:

rm -rf ~/Desktop/ai-video-editor ~/.ai-video-editor

Delete the evidence screenshots from the folder where you saved them.
Confirm that Finder no longer shows ai-video-editor on your Desktop or .ai-video-editor in your home folder.

Windows

Close Claude Code.
Remove the project folder and editor tool home by running this command:

Remove-Item -Recurse -Force "$HOME\Desktop\ai-video-editor", "$HOME\.ai-video-editor"

Delete the evidence screenshots from the folder where you saved them.
Confirm that File Explorer no longer shows ai-video-editor on your Desktop or .ai-video-editor in your user folder.

Linux

Close Claude Code.
Remove the project folder and editor tool home by running this command:

rm -rf ~/Desktop/ai-video-editor ~/.ai-video-editor

Delete the evidence screenshots from the folder where you saved them.
Confirm that your file manager no longer shows ai-video-editor on your Desktop or .ai-video-editor in your home folder.

Cleanup is complete when the private repository and both local folders are gone.

Nice Work!

That’s a wrap! With Claude Code, you transformed a raw talking-head video into a checked creator-style edit at edits/<name>/render.mp4.

You've learned how to:

Built a measured creator profile from five hand-picked videos with a shared editing style.
Reviewed a rendered first cut, combined timestamped corrections into one message, and approved the spoken story.
Rendered a creator-style video and saved one visual preference as a reusable taste rule.
Rendered the same edit on GitHub Actions from a private repo and downloaded it with github-fetch.

Ready to quiz yourself?