Video editors charge $250 to $1,000 per video; a Claude Code video editing pipeline reportedly costs about $27. The difference comes from free local tools and one paid Claude plan doing the coordination. Here is exactly how the seven-role system is built, trained, and run.
How Claude Code Video Editing Works
Claude Code video editing is an automated pipeline in which Claude Code, Anthropic agentic coding tool that runs in the terminal and on your local files, acts as the editor and project manager for a YouTube video. In an October 2026 walkthrough, creator Christian Peverelli showed footage edited entirely by this system, with every cut, sound effect, and animation produced without touching an editing timeline.
The system works because Claude Code runs locally: it reads your raw footage, launches other programs, and writes output to folders on your machine. It coordinates six other tools and follows a skill file that encodes your editing style. Your only jobs are recording the footage and giving feedback.
The Seven Roles in the Automated Pipeline
The walkthrough maps seven traditional video production jobs onto seven tools, each doing one role. Claude Code sits at the center and drives all of them.
| Role | Tool | What it does | Cost |
|---|---|---|---|
| Editor / coordinator | Claude Code | Reads footage, picks takes, drives every other tool | Claude subscription |
| Transcriber | Parakeet | Transcribes speech with timestamps, free, built by NVIDIA | Free |
| Animator | Hyperframes | Turns Claude-written HTML-style animations into 4K clips; credited in the video to HeyGen | Free |
| Researcher | Google Chrome | Grabs screenshots and recordings of posts, papers, and pages as B-roll | Free |
| Screen editor | Tella | Records face and screen; connects to Claude Code through an MCP connector | From $13/month (annual, as reported) |
| Sound designer | Epidemic Sound | Music and sound effects, searched and selected by Claude.99/month annual, as reported | |
| Finisher | FFmpeg | Cuts black frames, checks audio sync, renders the final file; free and over 20 years old | Free |
Two details make the pipeline practical. First, Parakeet runs locally, so Claude reads exact timestamps of what you said and finds your retakes. Second, animations are written by Claude tiny websites that Hyperframes renders at 4K, so graphics stay sharp on large screens without a separate After Effects seat.
How to Set Up the System Step by Step
Setup takes one Claude session and a handful of connections. The steps below follow the walkthrough's sequence for the Claude desktop app.
- Download the Claude desktop app from Anthropic sign in. The system needs a paid plan; the video recommends starting on the $20/month tier.
- OpenAI Claude Code session, switch it to local mode, and attach the working folder where your footage and outputs will live.
- Set permissions to auto so the agent can run commands without approving each one, and select a current model for the setup phase.
- In Settings, under Claude Code, enable web browsing.
- Prompt it: create a video editing system using Hyperframes, FFmpeg, Tella, and Epidemic Sound, and walk through the setup step by step.
- In Customize, open Connectors and add Tella and Epidemic Sound; log in through the browser when asked.
- If Hyperframes is not found automatically, locate its repository through a web search or the connectors list.
- Ask Claude to save the whole configuration as a skill, then OpenAI new thread and run a test edit.
- Supabase and other backend services mentioned in the walkthrough are optional add-ons, not requirements for the core editing loop.
The install itself is quick. What takes time, according to the video, is fine-tuning the look and feel, which is a training problem rather than a setup problem.
Training the Skill File: Teaching It Your Style
A skill is a plain file of instructions that Claude reads before doing the job. The creator's skill, called YouTube-Edit, runs about 1,400 lines and holds 36 rules covering colors, fonts, animations, and sound choices. Claude Code wrote the initial version itself; the human only described what good looks like.
Training follows a repeatable loop. Start with short clips, usually just a video hook. Give time-stamped feedback on each edit, for example: at second 22, make this element larger or change the color palette. Then close the loop with one instruction: save what you learned to the skill. The next session inherits every correction, and the walkthrough reports that repeated mistakes stopped appearing once a rule was saved.
Sound took the most iteration. On one hook, the creator auditioned eight to ten sound effects before finding a match, then pushed the volume of pops and clicks down until they felt right. Suspense in the hook followed by very quiet lo-fi hip-hop became a stored rule.
Running a Full Edit in Practice
A run looks like this: drop the raw file into the working folder, then prompt Claude to apply the YouTube editing skill. In the demo, the agent copied the clip, transcribed it with Parakeet, cut a take that stalled mid-sentence, and produced a 31-second hook from 2 minutes 46 seconds of raw footage, a process the creator says takes 10 to 20 minutes.
The output is reviewed like a contractor's first draft. Feedback comes as time-stamped notes on pacing, sound, and graphics, and any fix is folded back into the skill. For full videos, the creator now dumps all footage at once, including Tella recordings referenced by filename since Claude direct access. One batch ran overnight, and the finished video was ready to review the next morning.
What the System Actually Costs Per Video
All figures below are the creator's own accounting from the October 2026 walkthrough, not independent verification. He runs Claude on the Max plan at $200/month, but uses it for all his work; asking Claude to measure its own token use, he calculates a last video cost $24 in tokens at most.
The stack adds roughly $3 more per video: Tella at $13/month on an annual plan (about $2 per video at his output rate) and Epidemic Sound at $9.99/month annual. Parakeet, Hyperframes, and FFmpeg are free. His total: about $27 per video, compared with the $250 to $1,000+ per video he quotes for human YouTube editors.
The larger gain he reports is speed: planning, shooting, editing, and releasing on the same day, versus a three-to-seven-day turnaround with a human editor and multiple revision rounds.
Limits and Things to Know Before You Build This
The system has real boundaries. A paid Claude plan is required, and the setup phase burns tokens on a stronger model before cheaper runs. Hyperframes is a young free tool, so installation may need manual repository hunting. The workflow depends on local file access and connectors, which means your machine does the rendering work.
Cost claims are the creator's self-report. Token pricing on subscription plans changes, and $24 per video reflects his video length and plan limits, not a universal figure. Treat the $250-to-$1,000 editor comparison as his market framing rather than an audited benchmark.
The skill file is the actual product. Without trained rules for your fonts, colors, pacing, and sound, output will look generic. Budget several short sessions of time-stamped feedback before the edits match your taste. Communities that follow the creator's work, including groups such as Dev Doido do canal do youtube, share variations of this kind of setup, and you can also download his skill from his free We Are No Code community.
FAQ
- Do I need to know how to code to use Claude Code for video editing? No. The walkthrough's creator is not a programmer. Claude Code writes the scripts, runs the tools, and built the initial skill file; you provide footage and feedback.
- Is the transcription done in the cloud? No. The system uses NVIDIA's Parakeet speech recognition locally, which is why Claude read timestamps from files on your machine without uploading footage.
- How long does one automated edit take? The creator reports 10 to 20 minutes for a short hook and an overnight run for a full video. You then review and give time-stamped notes before the final version.
- What is the cheapest way to start? The video recommends the $20/month Claude plan plus the free tools. Tella and Epidemic Sound subscriptions can be added later or replaced with alternatives, though the connectors would change.
- Does the skill file improve over time? Yes, if you close the loop. After each feedback round, tell Claude to save the lessons to the skill, so every new edit inherits the corrections. The creator's skill reached 36 rules this way.
- Can it create B-roll for claims I make on camera? Yes. Claude uses a browser to capture screenshots or scrolling recordings of posts, papers, and pages, so the proof for a claim appears on screen.
- Who owns the tools in this pipeline? They are separate products: Claude Code and Claude from Anthropic, Parakeet from NVIDIA, Hyperframes credited in the video to HeyGen, FFmpeg as an independent open-source project, plus Tella and Epidemic Sound as commercial services.
- Does this replace a human editor entirely? Not in this workflow. The creator describes himself as the director: he records, reviews every edit, and gives feedback. The system removes timeline work, not creative judgment.
- What happens if the raw footage is messy? That is the normal case. In the demo, the agent handled a take where the speaker stopped mid-sentence to check notes, cutting around it automatically.
From Video Knowledge to Written Knowledge
This walkthrough worked because one creator documented his exact pipeline, costs, and training loop on camera. If you have similar hard-won knowledge sitting in your own YouTube videos, that same material can become a searchable written article instead of a 17-minute watch.
Skala blog turns a YouTube URL into a structured article: paste the link, transcribe the video, and generate a draft you can edit and publish. Some creators also mirror their guides on resource hubs such as crazystack.com.br to reach more readers.
Fork this article
Start a new branch from the same video, shaped your way. You keep the credit; the original keeps the attribution.
A fork in another language is filed as a translation of this article, so the two pages point at each other. You can unlink it later from the editor.
0/240
You are creating
- Format
- For
- Language
- Source
- Your angle
You will be asked to sign in before it is generated.
Buy credits