08:59Luuk Alleman · Demo
AI Just Ended Manual Video Editing (It's Better, Faster and More Creative)
A creator live-builds an entire YouTube video inside a Claude Code chat, then sells the system that did it.
July 16thThe argument in one line.
A single Claude Code skill can turn one long raw recording into a fully produced vertical short by handling transcription, topic selection, cutting, b-roll, captions, sound design, and color, leaving only cut-point correction to the human editor.
Read if. Skip if.
READ IF YOU ARE…
- You record long-form talking-head content and want to repurpose it into shorts without hiring an editor.
- You already use Claude Code or Claude Skills and want a concrete example of a production-grade automation skill.
- You're comfortable using DaVinci Resolve as a manual-correction step inside an otherwise automated pipeline.
SKIP IF…
- You want a fully hands-off, zero-review editing pipeline, since this workflow still requires a manual Resolve pass to fix cut points.
- You don't shoot enough raw footage to need topic-mining from a single long take.
The full version, fast.
The video demonstrates a Claude Code skill that automates short-form video editing from raw footage. The creator records one long take covering multiple topics, then asks Claude to identify short-worthy segments, transcribe the audio, and cut a vertical timeline that opens directly in DaVinci Resolve. He manually corrects a few imperfect cut points in Resolve, then hands the approved cut back to Claude, which plans and applies b-roll, captions, sound design, color grading, and music, and renders the final video. The skill's logic lives in a SKILL.md file plus a folder of Python scripts covering transcription, recutting, timeline alignment, and re-syncing sound. The creator positions this as close to full automation, with human review limited to correcting AI cut-point choices rather than doing any editing by hand.
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 →Where the time goes.
00:00 – 00:22
01 · Cold open
Creator claims the intro itself was fully AI-edited after 200+ hours building the skill.
00:22 – 00:59
02 · The raw footage problem
Explains the source is one long rambling recording he wants mined for about 10 short candidates.
00:59 – 02:21
03 · Prompting the skill
Shows the exact prompt given to Claude Code: identify shorts, cut one, output a vertical timeline openable in Resolve.
02:21 – 02:45
04 · Claude runs the cut
Claude transcribes, maps topics, identifies short candidates, and produces an aligned vertical timeline.
02:45 – 03:04
05 · Reviewing in Resolve
Opens the AI-cut timeline in DaVinci Resolve, plays it back to check quality.
03:04 – 03:37
06 · API key leak tangent
Tells the story of a leaked Google API key costing him $2,500, used as one of the short-form topic candidates.
03:37 – 04:11
07 · Manual cut-point correction
Fine-tunes a few cut points in Resolve where the AI cut was not quite right, then saves.
04:11 – 04:46
08 · Inside the skill files
Shows the scripts folder and SKILL.md that encode transcription, recutting, tail recovery, and timeline alignment logic.
04:46 – 05:37
09 · Finishing pass
Hands the corrected cut back to Claude to add b-roll, captions, sound design, color, and music automatically.
05:37 – 06:13
10 · First rendered result
Shows the finished first-pass render, calling it high quality for a first try and notes further fine-tuning is possible.
06:13 – 06:31
11 · Outro credits
Closes on a text card crediting Claude for edited-by, A-roll cut, B-roll, captions, sound design, color, music, and rendered-by; only filmed-by stays human.
Lines worth screenshotting.
- One long raw talking-head recording can be mined by an AI skill for multiple distinct short-form topics instead of scripting each short separately.
- The AI cutting is not perfect yet: the workflow still routes the auto-cut timeline through DaVinci Resolve for a manual correction pass before finishing.
- Once cut points are approved, b-roll placement, captions, sound design, color, and music can all be automated by the same skill in one pass.
- A production skill can be built from a SKILL.md file plus a folder of task-specific Python scripts for transcribing, recutting, recovering, and aligning footage.
- Leaked API keys are a real cost risk: the creator describes a $2,500 Google Cloud bill from a leaked key discovered while on vacation.
- Framing a workflow demo as claiming video editing is solved is itself a hook technique worth noting, even though the creator admits a manual correction step remains.
Repurposing long footage into shorts can be pipelined, not scripted one by one.
WHAT TO LEARN
Treating one long recording as raw material for an AI-driven topic-mining and cutting pipeline turns hours of unedited footage into a repeatable shorts engine, as long as a human still checks the cut points.
- A single long, unscripted recording can contain many distinct short-form topics that an AI transcript-and-topic-mapping pass can surface without manual scrubbing.
- Full automation is not reliable yet for the actual cut points; building in a manual review step, here DaVinci Resolve, keeps quality high while still saving most of the editing time.
- Once cut points are locked, the remaining production work such as b-roll, captions, sound design, color, and music is far more mechanical and automatable than the creative cutting decision itself.
- Documenting a repeatable process as a written skill file plus supporting scripts makes it reusable across many videos instead of a one-off trick.
- API keys leaking is a real and costly risk; a single leaked key generated an unexpected bill in the thousands of dollars, a reminder to rotate and scope keys carefully.
Terms worth knowing.
- Claude Skill
- A packaged set of instructions and scripts, a SKILL.md file plus supporting code, that teaches an AI coding agent how to perform a specific repeatable task such as video editing.
- A-roll / B-roll
- A-roll is the primary talking-head footage; B-roll is supplementary cutaway footage layered in to illustrate or break up the main shot.
- Timeline alignment
- The process of matching a re-cut or shortened video back to the original footage's timestamps so edits, captions, and sound stay in sync.
Things they pointed at.
02:45toolDaVinci Resolve
Lines you could clip.
00:00
“This intro was fully edited by my cloth scale, and I spent over two hundred hours building it.”
03:05
“I recently got a bill of two and a half k from Google, and I was not having any usage. My API key got leaked somewhere.”
06:26
“Filmed by Luuk. Edited by Claude. A-roll cut, B-roll, captions, sound design, color, music, rendered by Claude.”
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.
story
This intro was fully edited by my cloth scale, and I spent over two hundred hours building it, testing all the different strategies, testing all the different systems, and basically every single trick that I could find. And right now, this cloth scale is really saving me a lot of hours and money.
And in this video, I will show you the exact workflow. Step by step, super realistic. So let's go.
This is a pretty long video of me just yapping, like, all different kind of subjects. I can probably make around 10 short form videos from this. So what I want is I want Claude to basically get all of the text out of this video, find out what kind of shorts we can make for this.
Let me very shortly show you what that actually sounds like. It just Getsakes. So much more complexity in my app that stuff just started to break.
So that is one part from this whole video. As you can see, there are pauses in there, there are retakes in there, there are mess ups in there. This is actually something realistic.
So we have our cloth code right here. What I can do is I can paste the name of the actual video. So I say in downloads, I want to identify what shorts we can make from this, use our video edit skill.
For this, I want to just start with one short, please. First, just start with the cutting.
I then want to be able to open it in Resolve.
I will share more about this later on in the video so I can control the cuts that are being made. It should be a vertical video.
So this is all I do. As you can see right here, it really understands exactly what we should do here. So it is transcribing it.
It is then mapping the content, identifying the shorts, cut one, and hand us a vertical defined sheet timeline that we can then open in Resolve. So, it is currently running. Let's wait for that to be completed.
So, very quickly, if you're enjoying this video, make sure to like and subscribe. It really helps me to grow my channel, and way more content like this is in the pipeline. So, let's go on with the video.
So, as you see right here, it has all the topics of all the videos, all the things that I discussed in the video. And I could then very simply just click the one that I wanted, but this is something we could do for one video, 10 videos, or 50 videos. It is now working on it.
Let me now show you exactly what it looks like when we see it in Resolve. Because then we can just change whatever parts we don't like, push it back to cloth, and cloth will do all the finishing touches. Okay.
Perfect. So it was finished. As you can see, we already have the video right here, but this is not a video that we can edit already.
So what we need to do is we need to give Claude access to my resolve. So I click on allow for this session. We now opened our project in resolve.
And as you can see, when I play it, it looks like this. I am honestly very ashamed to say this, but I recently got a bill of 2 and a half k from Google, and I was not having any usage. My API key got leaked somewhere because honestly, I was just too ignorant.
With AI, it's now So as you can see, we can now edit the parts that are not perfect yet. As it can do a lot, but this is the one thing that doesn't really go perfect. So what we'll now do is the following.
We can now do it in our editor like this and make sure that our whole video is cut up in a way that works for us. It's free, by the way. Perfect.
I made a few adjustments to the video, some cuts that needed to be fine tuned. Right now, it's perfect.
So I want you to get that version and then start planning out all the b roll and all the other stuff. So we run that. Clot is actually doing everything for me.
It can handle the files, it can handle everything, so I don't really need to do anything. The interface is purely for editing and making it easy for me to check the video, make some adjustments, etcetera. Right here, I have my video cut aligned.
So, in our scale, we have the scripts and the skill dot m d. And our scripts basically contain Python files. And those are all part of the bigger plan.
So we transcribe, we recut, we recover some tails based on the sound waves, we make the timeline, we align stuff, do all these things that basically a real editor also does.
And then we have the skill dot m d. This is something I will share with you. Make sure to check the link in the description.
But that basically contains all my rules, hundred hours of fine tuning that I put into this skill. It explains the whole mental process of how to approach a video like this. And also, it says exactly when to use what Python script for what.
So as you see, it is a really large file containing all these commands, all these rules, as you can see here, the lessons, so everything that I learned over time. So I just added an image of me and my girlfriend in Japan.
It will now make the actual edit. So let me show you exactly what the first version looks like. I am honestly very ashamed to say this, but I recently got a bill of 2 and a half k from from Google, and I was not having any usage.
My API key got leaked somewhere because honestly, I was just too ignorant. With AI, it's now so simple to just create something and it works and your API keys are all over the place. And it is going well until it's not going well.
And when I saw that invoice, I was actually in Japan on my holiday. I I was sweating so hard. And honestly, my real reaction was it had to happen.
Because I was just not taking it serious enough. So please learn from me. Be cautious with what you do.
So this is very good and really for a first try, like the intro that you saw at the start of this video, that is really the highest quality it gets. I can now fine tune this way further.
I can now pick other visuals. I can even completely replace some visuals with others if I want.
30,000 businesses in The US right now that are a perfect fit for an AI employee layer, not tools, not automated. How does it even know what better means? Because get that run, and you've got a machine that confidently breaks a $100,000,000 earned by two people with AI.
And everyone started copying their stack, telehealth startup. Two person team, they use Cheti. Name is Luke.
And I dropped out of my AI degree in my final year. I just really wanted to build real stuff rather than just learning.
Thank you for sticking around until the end of this video. Make sure to check the links in the description. Make sure to like and subscribe, and then we'll see each other in my next video.
The Hook
The bait, then the rug-pull.
A creator claims two hundred hours of trial and error produced a Claude Code skill that can transcribe a raw recording, find the shorts hiding inside it, cut them, and finish them end to end.
Named ideas worth stealing.
01:40model
Raw-take-to-shorts pipeline
- Record one long raw take
- AI transcribes and maps topics
- AI proposes short candidates
- AI cuts one short as a vertical timeline
- Human corrects cut points in Resolve
- AI finishes b-roll, captions, sound, color, music
- Render
A repeatable pipeline for turning a single long recording into multiple finished shorts with AI handling every step except cut-point correction.
Steal forAny creator sitting on unedited long-form footage who wants a repeatable shorts pipeline
How they asked for the click.
VERBAL ASK
06:20link
“Get The Video Editing Skill for FREE / Learn How to Build AI Systems for FREE, links in the description”
Soft CTA delivered on-screen as a text card and reinforced verbally with a subscribe ask; both links are free lead magnets, not a direct paid pitch.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
