Big Idea

The argument in one line.

Writing your own AI editing skill tuned to your personal editing logic produces cleaner rough cuts than paying hundreds of dollars per year for a generic tool that still leaves gaps you fix by hand.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You film talking-head tutorials, courses, or vlogs and regularly spend 2-3 hours cleaning up dead space and bad takes.
You use Claude Code or want to start using it to automate repetitive production tasks.
You have been paying for Gling, Descript, or similar AI editors and are frustrated by the cleanup work they leave behind.
You have good mic quality and a quiet recording environment -- the skill requires clean audio to work well.

SKIP IF…

Your videos rely on B-roll, VFX, zoom-ins, or multi-camera setups -- this skill handles cuts only.
Your recording environment has significant background noise, which breaks the waveform-based cut detection.
You do not use Premiere, DaVinci Resolve, Avid, or Remotion -- the outputs are editing-app files, not a finished render.

TL;DR

The full version, fast.

Off-the-shelf AI editors transcribe your video and guess at cuts, leaving gap frames and bad takes you still clean up manually. Perfect Cuts works differently: it stores every spoken line in a CSV with a keep/cut decision and a reason, runs an LLM pass to confirm the full edit makes sense, then uses dB-threshold analysis to make the actual frame-level cuts tight. The output is a package of files -- XML for Premiere or DaVinci, SRT for captions, a Remotion browser editor, and a reviewable spreadsheet so no cut is permanent or hidden. A 40-minute raw course becomes a clean edit in roughly 12 minutes.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 01:08

01 · Hook + proof

Shows finished 40-min course cut in Premiere; claims Claude did it in under 10 min; sets up the demo

01:09 – 01:47

02 · Credibility + channel intro

Vic introduces himself -- 20 years editing, 5 years running channels, 650K followers across other channels

01:48 – 02:54

03 · How the skill works

Four-step pipeline: transcribe, line-by-line CSV with keep/cut decisions, LLM coherence review, dB-threshold waveform cuts

02:54 – 03:15

04 · Caveats

Requires clean mic; no VFX or zoom-ins; works on short-form too

03:15 – 04:26

05 · Live install demo

Fresh install of the skill zip via Claude Code chat on a PC; Claude handles dependency setup automatically

04:26 – 06:25

06 · Setup questions walkthrough

Skill asks script type (exact/rough/freestyle), whether to include rendered MP4, and output path

06:25 – 07:44

07 · Output package reveal

Shows README, XML timeline, CSV with cut reasons, Remotion launcher, SRT captions, EDL, and JSON cuts data

07:44 – 09:28

08 · Premiere demo

Loads XML into Premiere; plays the sequence; no gaps, tight cuts -- compares it to how a pro editor would do it

09:28 – 10:24

09 · Value recap + free offer

Quantifies time saved; announces free Skool community at skool.com/vic with skill download

10:24 – 11:37

10 · CTA / channel pitch

Promotes free Stay Human course; recaps channel stats; subscribe ask

11:37 – 11:47

11 · Outro

Channel logo end card with subscribe prompt

Atomic Insights

Lines worth screenshotting.

Generic AI editing tools leave gap frames because they were not trained on your specific editing logic -- they guess.
Storing cut decisions in a CSV with reasons turns a destructive auto-edit into a fully reviewable, reversible process.
A 40-minute raw talking-head course can be cut to a clean sequence in about 12 minutes with a well-tuned LLM skill.
Clean mic audio is not just a quality preference when using AI editing -- it is the technical prerequisite the dB-threshold step requires.
The LLM coherence review pass is what separates a rough cut from a sequence that actually makes sense -- the transcription alone is not enough.
Building a skill around your own editing logic costs nothing but time and produces output you would actually ship, unlike a generic yearly subscription.
Outputting to XML, SRT, EDL, and JSON simultaneously means the edit decisions work in any app and are never locked into one platform.
Reviewing a 40-minute timeline at 2x speed after an AI rough cut is faster than reviewing raw footage at 1x -- the skip-forward benefit compounds.
The script-type question (exact/rough/freestyle) is what allows the LLM to calibrate how aggressively to cut repeated or improvised lines.
A Remotion launcher in the output package means creators without Premiere or DaVinci can still work with the edit in a browser.

Takeaway

A skill you tune once does the boring edit.

WHAT TO LEARN

Off-the-shelf AI editors fail because they do not know your editing logic -- a thin custom skill that encodes your own rules produces cuts you would actually ship.

Generic AI editing tools produce rough cuts with gap frames you still fix by hand -- the value disappears in the cleanup time.
A four-step pipeline (transcribe, CSV, LLM review, dB cuts) replicates how a professional editor actually thinks about a talking-head timeline.
Storing cut decisions in a reviewable spreadsheet with reasons makes the process non-destructive: find any cut line, see why it was removed, and reverse it without digging through raw footage.
Clean audio is the only hard prerequisite -- background noise breaks the dB-threshold step, making mic quality a technical constraint, not just a preference.
Outputting to XML, SRT, EDL, and Remotion simultaneously means the edit decisions work in any app and are never locked into one platform.
A 40-minute raw shoot becomes a fully cut timeline in about 12 minutes, converting hours of scrubbing into a single review pass at 2x speed.

Glossary

Terms worth knowing.

Perfect Cuts: A custom Claude Code skill that automates the rough-cut editing of talking-head video by transcribing, analyzing, and slicing a raw clip based on LLM decisions and audio waveform thresholds.
dB-threshold cut: A cut point determined by measuring audio decibel levels; silence below the threshold marks gaps between phrases, allowing frame-accurate cuts without manually finding the exact frame.
Remotion: An open-source JavaScript framework that lets you define and render video in a browser environment, used here as a no-install editing interface for the cut sequence.
EDL (Edit Decision List): A plain-text file format that stores edit decisions as in/out timecodes, compatible with Avid and older editing applications.
Gling: A subscription-based AI video editing tool that auto-removes silences and bad takes from talking-head footage; the paid tool this skill was built to replace.

Resources

Things they pointed at.

01:48productGling ↗

06:25toolRemotion ↗

09:28linkAI Systems for Human Content (Skool) ↗

10:24productStay Human (free course) ↗

Quotables

Lines you could clip.

00:48

“I felt like if I made my own version of this skill, I could tune it in a way that follows the same logic I use when editing any of my own videos.”

Core insight in one sentence -- the case for custom over generic→ TikTok hook↗ Tweet quote

09:12

“Look at how clean these cuts are. Like, there are no gaps between anything. It is exactly how a pro editor would do it.”

Payoff moment with visual proof on screen -- ideal clip-and-react format→ IG reel cold open↗ Tweet quote

09:29

“This would have taken me hours to go through a forty minute course and edit. Now I can just watch the entire thing at double speed just to make sure.”

Concrete time-save claim with a specific workflow→ newsletter pull-quote↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

This video project has a forty minute runtime, but Claw just edited the entire thing for me in under ten minutes. It scanned the entire transcript and determined what to keep and what to cut. It then chopped all of the footage, removing all of the gaps, and also cutting it super tight so there wasn't a bunch of fluff between everything that I said.

Actually, editing the video based off of the waveforms like a pro. Now there are other Claude skills that will edit video for you just like this, but every single one that I downloaded, I found only gave me a very rough draft. And instead of having a clean-cut like this, there would often be a bunch of frames left over of dead space that I would have to manually go and fix after.

But since I personally have nearly twenty years of video editing experience, I felt like if I made my own version of this skill, I could tune it in a way that follows the same logic I use when editing any of my own videos. If you stick until the very end of this video, I'm gonna show you where you can download this Claude skill completely for free.

My name is Vic. I'm not an AI guru. I gotta make that super clear to you guys.

I'm a content creator who's been running my own channels for over five years. On my other channels, I have over 650,000 followers, and this entire channel here is to develop systems using AI that helps you make content and keep it human.

So if that sounds like something you're interested in learning, make sure you subscribe to the channel and turn on the notification bell. For the last year or so, I've used Gling, which is an AI tool, which does exactly what this Claude skill can do, but they charge hundreds of dollars a year, and it gives you a super, super rough cut.

So it transcribes a whole video, tries to pick the good and bad takes. Usually, it fails at it. And then when you actually look at the cuts, they're super wonky, and they have bunch of gaps between them.

And there's so much cleanup required with using a tool like this. And without Gleng or a different similar AI tool, going through a video like this and editing it can literally take, like, two or three hours.

So I was using Gling, and I was happy, but I just felt that there was a better way with Claude code. That's when I started looking around at different YouTube videos and different GitHub repos and finding what other skills were doing, and, essentially, I took all of that in and made my own version of a skill that can do it better.

First off, it transcribes the whole video, and then line by line stores it in a spreadsheet, which is how it decides which lines to keep and which lines to get rid of because if you're filming, you sometimes you repeat yourself. Then the LLM actually reads the final product to make sure it actually makes sense, and then it looks at the waveforms of the audio and makes those cuts even more precise based on a dB threshold.

So I will say a caveat to this skill is that your audio needs to be good, and you can't have a bunch of background noise because that will probably mess with how it actually works.

But if you have a good mic and you're in a silent environment, this thing will work extremely well. Also, I need to make it clear that this tool doesn't do VFX. It doesn't do zoom ins or anything like that.

All of that stuff is done afterwards. This tool was designed to cut and slice and remove all the crap from long stretches of video, but it can also work on like a one minute short form video or anything like that.

This is an entire course that I just filmed in one shot. It is forty minutes worth of footage. Okay?

Like, this would take me two to three hours to go through, cut out all of the parts where I do a bad take, all of the gaps, and really just clean up the entire timeline.

Okay? This won't take me that long. I've done it before.

And a forty minute video, that's quite a long time. So let's actually try this. Install the skill from scratch and then take this entire course that I just filmed and run it through the skill and see what we get.

We have the zip file right here, and I'm just gonna say install this skill for me that I made on my MacBook. Because I actually made it on my MacBook, and I'm bringing it over to my PC here.

So this is literally a clean, fresh install of this skill. So we'll see. I know it needs some dependencies.

It needs to install a few things, and it's gonna completely do that on its own in the background. See, it has an install text right here, and it's completely safe to run as you can see. But if you want, you can ask Claude to scan it for security as well if you feel like you need to.

Alright. He says it's installed and ready to go, and all we need to do is drop a raw talking head clip and say, cut this clip, or you can say, like, you know, slash perfect cuts, which is the actual skill name to start it up, and it's gonna guide us through the entire process.

So I'm just gonna paste the file path of the video that I just recorded, and then I'm going to say use the perfect cuts skill.

What it's going to do now is ask me a bunch of questions about my project. It realized once I ran it that I didn't have the right version of Python, so I'm just gonna get it to install the right version of Python. But then these are the questions that it will ask every single time.

Did you film from a script? This is super important because if you film a video, maybe a short form video, and it's like a minute long, and you go line by line script exactly how the script says, you can give it the script and it'll just be even more accurate. Or if you have a rough script and then you maybe freestyled and improved it as you recorded it, you can say I have a rough script or you just completely freestyled it.

So there's three options, exact script where it wants you to give it, rough script, it will also want it as well, and then freestyled, no script. I'm just gonna say that because this is a course, and I just literally yapped. So freestyled, include a rendered m p four in the package.

This will increase the file size of the package that you get at the end because the package is going to give you a ton of different file types. If you want that rendered MP four, once you start using the skill and you're like, damn, it's crushing it every time, you can say yes. This time for me, I'm going to say no because it's forty minutes long, and it's probably gonna be pretty big, and I wanna work with this file after.

So I'm gonna say no, but just know that it will render a a maximum quality m p four for you as well as these other workable files. And then it's gonna ask where should the package go. So I just grabbed a file path of exactly where I want this to be saved, and then I'm going to submit.

And now this is going to work with Claude, and we'll be back once it's done to take a look at our results. A few moments later. Alright.

Alright. She is completed. It took about twelve minutes to run.

However, I actually built this on my MacBook, and then I just transferred it over to my PC. So it just had a bit of rewiring to do. So let's take a look at what the package has.

So here it is right here. We have a readme file, which I'll open first. So it basically outlines what each file will do for you.

It also gives me credit. Hello. So the first one is an XML file.

This is an editing file for Premiere and DaVinci Resolve. The second one is a revive file.

So this is all of the cut decisions in a CSV spreadsheet. I'll show you what that looks like.

This is so if it makes a cut that you didn't want, and you're like, oh, it it cut that line out. Now I have to go dig through all the footage. No.

No. No. You guys just open the spreadsheet, find the line, and then work off of that point.

Then, also, it creates a file that you can double click, and it will open the Remotion window on your browser if you don't have Premiere or Resolve, so you can edit in Remotion built right in. Then we have a captions SRT ready to upload.

So let's say you did use the m p four and you just wanted to go right to YouTube, you have the captions file that you can upload as well. Then you have a EDL file, which is very similar to the XML. It's just other different editing softwares like Avid or Legacy.

And then we have a cuts data file, which is a JSON file. So first off, let's take a look at the Remotion file. I'm just gonna double click this Windows Remotion file right here.

And here it is. We have the entire video cut in Remotion. I can edit it right on my browser if I don't have Premiere or Resolve or anything like that installed.

And here is the spreadsheet file, the CSV, with all of the text lines in it. So you can see right here, this is the entire script line by line, and then it's the status of kept or cut.

So it went through and decided, like, okay. Which ones are we keeping? Which ones are we cutting?

Now this is super important because if you want to go back later and find a specific line that it might have cut, you can find it and you can see exactly what happened here. And there's also even a reason on why it decided to cut it. So right here, false started, restarted at forty three.

And this is just so it's not as destructive, and you can have more control later on after the first edit pass is done. This is the original Adobe Premiere file, which I was in before, and I can just drag in the XML file right into that project.

And then here's the sequence that it created. So this is my first time looking at it too. Let's double click and open up the sequence.

And you can see, boom, bro. This entire thing has been edited from start to finish for us. And look at how clean these cuts are.

Like, there's no gaps between anything. It's exactly how a pro editor would do it. So let's actually, like, listen to the alright.

So let's, like, start this from the start and see how it sounds. Have you ever heard of the dead Internet theory? It's basically the theory that bots have taken over the Internet and contribute to, like Like, it's so clean.

There's no gaps between anything. Everything's super cohesive. This would have taken me hours to go through a forty minute course and edit.

Now I can just watch the entire thing at double speed or even triple speed just to make sure, and I just do that by clicking l on my keyboard twice. If I wanted to go over it all, which I I will, I'm still gonna manually scrape through it all, but you can see this is a crazy tool that saves you hours and hours and hours.

And like I promised at the beginning of the video, I'm gonna give you this skill completely for free. I'm also going to give you access to that course that I was editing completely for free as well.

And that's all in my free school community called AI Systems for Human Content. This school community is just over a week old. We already have 1,300 members in it, and I'm dropping new skills in here all the time and also free courses as you can see.

So once you're here, go over to classroom at the top, and the free skill will be located in AI automations. It's not uploaded just yet because I'm in the middle of filming this video, but there will be a perfect cuts video editor skill here where you can click that and then grab the file. This is from my last video, my Higgs field replacement.

And there will also be a another block that will have the free course. That course is called stay human. It's all about how to create content with AI while keeping your content human.

If you want to start a business through content on YouTube, on Instagram, on TikTok, whatever platform you want, that course will help you cut through a ton of the noise. Like I mentioned, I've been a professional creator for five years. I've been editing video for twenty years.

I've made millions online even though people think this is my only channel. It's not. I have 650,000 subscribers and followers across all of my other channels.

So this is the business that I run. This channel that I'm on right now, this is brand new. This is, like, a week and a half old at its current time, and I have 1,500 followers.

And the last video that I dropped right now, like, literally, there's three videos on this channel. This video is currently sitting at 27,000 views.

So clearly, I know what I'm doing. If you got some value from this video, subscribe to this channel and turn on notification bells. My name is Vic, and make sure you stay human.

The Hook

The bait, then the rug-pull.

Vic opens mid-Premiere, timeline already cut, and makes the claim before he introduces himself: a 40-minute raw course, edited to a clean sequence in under 10 minutes by a Claude Code skill he built himself. The promise is not AI editing in general -- it is AI editing that works the way a professional editor would do it.

Frameworks

Named ideas worth stealing.

01:48list

Perfect Cuts 4-Step Pipeline

Transcribe full video
Store line-by-line in CSV with keep/cut decision + reason
LLM reads full output to verify coherence
dB-threshold waveform analysis for frame-accurate cuts

The four ordered steps the skill runs to turn a raw talking-head clip into a clean cut sequence with no gap frames.

Steal forAny batch-processing workflow that needs human-readable audit trails alongside automated decisions

06:25list

Output Package Formats

XML (Premiere/DaVinci import)
CSV (reviewable cut decisions with reasons)
Remotion launcher (browser-based editing)
SRT (captions ready to upload)
EDL (Avid/legacy apps)
JSON cuts data (raw machine-readable)

The six file types the skill produces so the edit decisions are usable in any editing environment and fully reversible.

Steal forAny AI workflow where you want the output to be portable and non-destructive

CTA Breakdown

How they asked for the click.

VERBAL ASK

09:28link

“I am going to give you this skill completely for free. That is all in my free school community called AI Systems for Human Content.”

Soft lead-in at 9:28 with the payoff promise from the cold open, then explicit Skool link at 10:00. Subscribe ask layered on top. Free course bundled in as a second incentive. Well-executed double CTA.

MENTIONED ON CAMERA

01:48productGling ↗

06:25toolRemotion ↗

09:28linkAI Systems for Human Content (Skool) ↗

10:24productStay Human (free course) ↗

FROM THE DESCRIPTION

PRIMARY CTAWhere the creator wants you to go next.