YT YouTube2Text

YouTube2Text → Transcripts → I Fully Automated My Video Editing Using Claude Code (Full Walkthrough)

I Fully Automated My Video Editing Using Claude Code (Full Walkthrough) — Transcript

by Christian Peverelli · 3,799 words · 523 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Everything you're seeing in this video
0:01was edited by my Claude Code system
0:04while I was asleep. Every cut, every
0:06sound, every animation like this one
0:09>> [music]
0:10>> or this one
0:12or this one
0:15was made without touching a timeline,
0:17and no, I don't know how to code. So,
0:19today I'm going to pull the curtain and
0:20show you exactly how I built this expert
0:22editor step by step. The tools, how to
0:25set them up, how I taught it my style,
0:27the exact cost, and how I'm saving
0:29thousands of dollars a month. Let's go.
0:31Okay, so making a video like this one it
0:33used to take a small team. And to be
0:35clear, I still have one job to hit
0:37record and also to provide feedback. But
0:39in my system, every other job goes to a
0:41different tool. So, I'm going to break
0:42these down for you one by one. Now, the
0:44first one is the editor. That's the
0:46person who sits in front of the timeline
0:48for hours. They watch every clip, they
0:51pick the ones they like, cut out the
0:52ones that they don't, decides where
0:54every zoom goes, and also figures out
0:56the graphics and animations that they
0:58want and where. Now, in my system,
0:59Claude Code is the tool that does all of
1:01that. Think of this role like the
1:03project manager of the entire video.
1:05Now, I personally use the desktop
1:07version of Claude. It works perfectly
1:09there. And Claude Code can work locally,
1:12which means that it actually uses my
1:14computer for me. It can access the
1:15files, it uses other programs, and
1:18finally it's able to put the output
1:19inside of a folder on my computer. Now,
1:21that's actually important because Claude
1:23actually is the coordinator of all the
1:25other tools here. But there's one catch
1:27here. Claude can't actually hear my
1:29videos. So, how does it know what I said
1:31and where I messed up to be able to cut
1:33the whole thing together? Now, that's
1:34the second job that we have in this
1:36list. It is the job of the transcriber.
1:38Now, this is the person who listens to
1:39all the footage, transforms that, types
1:41it out into text, and puts timestamps on
1:44these. And the tool that does that is
1:46called Parakeet. It's completely free
1:48tool, and it was built by Nvidia, which
1:50is the leading manufacturer of chips.
1:52No, not Doritos or Lays. I mean like the
1:54chips, right, for AI. Mm, chips. Get
1:58back to it. So Parakeet does the
1:59transcription and then Claude is able to
2:01actually read the timestamps and exactly
2:04what is being said and is able to then
2:05use that for the next step of the
2:07process. This is how we're able to
2:08identify and chop up the footage so that
2:10we have the right takes. Job three is
2:12actually the animator. This is the
2:14person who builds every single title,
2:16every logo that pops in, any motion
2:19graphics. And in the old world, this was
2:21done with After Effects. You literally
2:22had to use one tool just to do these
2:24things. And this is exactly where
2:26Hyperframes comes in, tool number three.
2:28Now this again is a free tool developed
2:30by HeyGen, which is the video platform.
2:33Claude writes each animation like a tiny
2:35website and then Hyperframe turns that
2:38into a 4K video. So now the animations
2:41can look sharp even on a large TV. It
2:43all runs on my own computer so it
2:44actually costs me nothing. However, in
2:46videos sometimes I'll say things like
2:48OpenAI just announced blah blah blah and
2:50when I say that, I actually need the
2:53fourth person which is the researcher.
2:55Someone who goes out and is able to find
2:57a screenshot of that announcement or a
2:59post on X or a research paper that
3:02proves the point that I'm actually
3:03illustrating and turns that into what we
3:05call B-roll, things that you show on top
3:07of the video that are basically outside
3:09footage. And the thing that does that is
3:11something we all actually have already
3:12on our laptops which is a browser.
3:14[music] So in my case, I use Google
3:16Chrome and then Claude is able to
3:18actually use the browser to go out there
3:20and do that research. And in the
3:21background, Claude, my researcher, is
3:23able to grab screenshots or record video
3:26scrolling down a website and then it
3:28could do things like highlight lines or
3:30actually zoom into specific sections of
3:32those pages. And the great thing is that
3:34it only shows you exact facts. So it's
3:36really taking the original X post or the
3:39actual webpage. So I can just say what
3:41happens and the proof shows up on
3:42screen. Now the fifth job is the screen
3:44editor. Now this is mainly because I
3:46love making simple tutorials teaching
3:48people how to use AI tools, how to make
3:50money with them, but I use screen
3:52sharing a lot, and that's why I used a
3:53tool called Tella. Tella records my face
3:56and my screen at the same time, and
3:58because Tella has a connector called an
4:00MCP, I can connect it directly into
4:02Claude Code, and it's able to do all the
4:04things it does inside of the software.
4:07The cuts, the zooms, the layout, the
4:09formatting, yada yada, everything that
4:10you can do inside the software, you're
4:12able to do via Claude Code. So, I
4:14literally only just have to record
4:16through it, and it does the rest. But,
4:18up until now, we've just been talking
4:19about the visuals of this video. Every
4:21great video team has what we call a
4:23sound designer. This is a person who
4:25decides what music goes on each clip,
4:27what sound effects go in when specific
4:30things happen, and make sure that these
4:31are also on brand. Because you probably
4:34don't want music styles that make you
4:35want to sleep when you're teaching a
4:37tutorial, for example. So, for this, I
4:38have a sound base called Epidemic Sound.
4:41It is something I pay for, we'll talk
4:42about pricing at the end, and I'm also
4:44able to connect it to Claude, and it's
4:46going to be able to go out, search for
4:48sounds, and the cool thing is that I can
4:49train it on what I like. Usually, I like
4:51having the hook be a little bit
4:52suspenseful, and then I go into some
4:54kind of upbeat music that's kind of like
4:56lo-fi hip-hop, and have it very, very
4:58low. And so, yeah, sound effects also
5:00have their own rules. I'll talk about
5:01how I set up all of this right after.
5:03And now, the seventh member of the team
5:05is actually the finisher. This is the
5:07person who joins all the pieces together
5:08and looks to see if there's no mistakes.
5:10This tool is called FFmpeg. It's free,
5:13and it's been cutting videos for more
5:14than 20 years. In fact, it's under the
5:17hood of pretty much every editing
5:18software that exists today. So, Claude
5:20uses it, and it's responsible for
5:21cutting out any black screens, checks if
5:24there's nothing off with audio sync
5:25before I ever get to even see the video.
5:28So, those are the seven tools or jobs
5:29that have to be done, but none of those
5:31actually know what my video's supposed
5:33to look like. That's probably the most
5:34important thing, and it's called the
5:36skill. A skill is simply just a file of
5:39instructions that Claude reads before it
5:41does the job, and it's pretty much my
5:43editing style broken down into
5:45specifics. The colors, the fonts, the
5:48animations, the sounds, with a bunch of
5:50details that are basically my
5:51preferences. Mine is called YouTube-Edit
5:53and it's about 1,400 lines, but the
5:56great thing is that Claude Code actually
5:58built this whole thing for me. So,
5:59pretty much Claude runs all of the tools
6:01and then I tell Claude what good looks
6:03like. Now, I'm not technical and you
6:04probably aren't either, so I'm going to
6:06show you exactly how to set this up.
6:07It's super easy. Okay, so the first step
6:09to set this up is to download Claude.
6:11Now, you can go to Anthropic's website
6:13and you can download the desktop app.
6:15That's what I recommend. It's what I
6:17use. And once you've properly installed
6:19that, it's going to ask you to create an
6:20account. You're going to have to use
6:22either Google or your username and
6:23password to do that. And if it is the
6:25first time you download Claude in
6:26general, it will try to set you up. I do
6:29recommend you go through that. I have a
6:31full other video on how to get set up
6:33properly. You don't actually need that
6:34here. Now, for this you do need to have
6:36a paid plan. I would upgrade to the $20
6:38plan. No need to go for higher than that
6:40at this point. And when you hit the main
6:42interface, it's going to look something
6:44like this. It might look a little
6:45different just because things change
6:46constantly. But basically, if you click
6:49on new, you're going to be able to start
6:50a new session inside of Claude. Now, up
6:54here on the left, we have Claude, which
6:55is the normal version, which is chat and
6:57co-work. And right here, we have Claude
6:59Code. So, for this, you want to be on
7:01Claude Code. And then, there are a
7:03couple setup steps you need to do before
7:04we actually move forward. First off, you
7:06want to make sure that you are on local,
7:09meaning that it's using your computer to
7:10do this. Next up, you need to actually
7:12determine the folder in which you're
7:14going to be working. For this, you can
7:16just click on this. As you can see, I'm
7:18already connected to my Claude Code
7:20folder, but you can basically click on
7:22this one and you can add a completely
7:23new folder. You can create one here and
7:26then you attach it. After that, you want
7:28to turn the permissions on auto so that
7:30it can run without you having to say yes
7:32no to thousands of things. And then,
7:34you're going to want to switch the model
7:35to one of the latest models just because
7:37the setup is an important step. After
7:39that, it will be able to run on a
7:40cheaper model. And finally, you do want
7:42to come down here to the left and you
7:44want to go inside of settings. You need
7:46to go to Claude Code and here you can
7:49copy the exact settings that I have here
7:51or you can just decide to basically turn
7:53on one very important thing for us to be
7:56able to browse the internet, which is to
7:58turn that on and that's done right here.
8:01When you're done, you can pretty much
8:03close this and you can go back to the
8:05main menu, click on you. Then Claude
8:07Code is actually going to be very good
8:08at setting this up itself. So in this
8:11case, I'll give it a prompt like this.
8:13Could you please create a video editing
8:15system for me? I'd like to use
8:17Hyperframes, FFmpeg, Tella, and Epidemic
8:21Sounds for this system. Run me through
8:23how to set it up step by step.
8:26So one route is simply to get it to
8:27guide you through the process. Now the
8:29other way to do this, Claude should be
8:30able to download and install everything
8:32that it needs and it can, it will ask
8:34you to actually connect a couple of
8:36these tools as well. You come here to
8:38the left and you go into customize
8:40>> [music]
8:40>> and you go into connectors. And then you
8:42can pretty much select the tool. So in
8:44this case, I'll go and look for Tella,
8:46which is my screen recording device. As
8:48you can see, I've already attached this.
8:50If not, you can just click on the plus
8:52button right there. And for Epidemic
8:54Sounds, it's the same thing. Look for
8:55Epidemic Sounds.
8:57>> [music]
8:57>> There you go, it is the music and sound
8:58effects and you can go ahead and turn
9:00this on. It'll ask you to log in in the
9:02separate browser. Once you've done that,
9:03you'll see this check mark. Now all the
9:05other things it needs, it'll download
9:06itself, but if it doesn't find
9:08Hyperframes for any reason, you can
9:10simply Google Hyperframes, you'll find
9:12the link to the repo, but you can also
9:14find Hyperframes inside of the
9:16connectors, too. Now once you've
9:17connected all those things, you want to
9:19make sure that it saves them as a skill
9:21on your computer. And once that has
9:23actually saved inside that one chat or
9:25thread, you want to open a new thread
9:27and give it a first test. Now as you can
9:29see, setting these things up is pretty
9:30straightforward. What takes time is to
9:32basically fine-tune the look and feel.
9:35And if you'd like to just download my
9:36skill and use that, you can feel free to
9:38do so inside my free community. Link is
9:40below. But now let's jump into how I
9:42actually taught it my style. Well, at
9:44first I pretty much gave it notes on
9:45every single thing I made, and I started
9:48with small clips, usually just the hook
9:50of a video. And when it gave me a first
9:52edit, I would basically give it very
9:53specific time-stamped feedback on the
9:56work it had done. And then 10 days in, I
9:58actually took one of the hooks and
9:59started just implementing sound. I had
10:02it try a bunch of different sound
10:03effects on one hook, and I had to go
10:05through about eight to 10 different
10:06sounds until I found something that
10:08really worked for what I enjoyed.
10:10Because what it had made was a little
10:11too busy for my liking. So, this was
10:13mainly used for sound effects, right?
10:15When something pops up, you could have a
10:16little pop. When something's clicked on,
10:18you can have the little clicking sound.
10:20And as it started making decisions on
10:22how to use the different sounds I gave
10:23it access to, I basically told it,
10:25"Okay, let's reduce the sound of that.
10:27Okay, I think that I don't like this
10:29sound at all. Just take it completely
10:31out." And slowly but surely, it was
10:33learning exactly what I liked. Now,
10:34here's the absolute key to all of this.
10:36When you actually give it feedback, and
10:38after that, you tell it to save what
10:40it's learned to the actual skill itself.
10:43That way, when you create a new edit,
10:45you'll constantly be improving your
10:46editing system. Then I created a new
10:48session where I had this updated skill,
10:50and I basically put another hook. This
10:52time, I loved that it had gone out and
10:54grabbed a specific logo and used the
10:56sound effect to show it, but I wanted it
10:58to be shown slightly larger. I also
11:00didn't really like the color palette it
11:02was using for some sections or some
11:04animations. So, I decided to switch it
11:06up a little bit. And the more videos I
11:07made and provided feedback on, the
11:09better it got. And today, my skill has
11:1136 different rules that it has to abide
11:13to. Then I started discovering really
11:15cool new animations, and I started
11:17deciding which ones would make sense for
11:18the content that I actually made. And
11:20so, slowly I started adding more of
11:22these. Every time I used them, I would
11:24fine-tune them. And awesome thing is
11:26that every note that I gave it, the next
11:28time it never made that same mistake
11:30again. Meaning that I basically was
11:32getting exactly the output that I was
11:33looking for. All right, my friends,
11:35enough talking. Let me actually show you
11:37this in action. Okay, so we're back
11:38inside Cloud Code now, and since I've
11:41already created the skill or the first
11:42version of it, I can pretty much run it
11:44at any time. I'll come here. I'm going
11:47to basically add a raw file. So here I
11:49have a hook that I recorded earlier
11:52today for this specific video. I'm going
11:55to go ahead and open this one. Then I'm
11:57going to tell it this.
11:59Please use my YouTube editing skill to
12:02edit this hook for my YouTube video. I
12:05can also choose to add any additional
12:06things I'd like if there are specific
12:08things I'd like it to do inside the
12:10hook. So in this case, when I say things
12:13like zoom, you can actually show a zoom.
12:15When I talk about a sound effect, you
12:18can make a sound effect. And then when I
12:20talk about the different animations, you
12:23can basically showcase three of the most
12:25impressive animations, either that I've
12:27already done in the past or that you
12:29really like from the library that we
12:31have inside the skill. So those are just
12:32instructions that I'm giving it because
12:34I know the hook I recorded and I'd like
12:36to actually show the editing skill
12:38inside the hook as well. But you can
12:39give it pretty much any instructions. At
12:41this point, I'm going to pretty much
12:43just press enter and let's see what
12:45Cloud actually does. Running the skill,
12:47YouTube edits as I said, so it finds the
12:49skill and it just starts running it.
12:51Probing the clip and listing, okay. Now
12:54just to be clear, this is a completely
12:56unedited thing. So basically I make lots
12:59of mistakes, I stop in the middle of the
13:01hook, and uh look at my notes. It is
13:03basically cutting it up, so it's copying
13:05the clip, transcribing it using
13:07Parakeet, which is the tool that we
13:09talked about earlier. Now, that will
13:11install automatically in the setup as we
13:12talked about. Now because this process
13:14will take anywhere from 10 to maybe 20
13:16minutes, I'll show you the V1 of the
13:18hook. Everything you're seeing in this
13:20video was edited by my Cloud Code system
13:23while I was asleep. Every cut, every
13:26sound, every animation like this one
13:29or this one
13:32or this one
13:34was made without touching a timeline and
13:36no, I don't know how to code. Boom. So,
13:38as you see, there's a couple things I
13:40would change from this, but it's pretty
13:41much spot-on. It basically says that it
13:44cut down 2 minutes and 46 of raw footage
13:46into 31 seconds, which is absolutely
13:48great. And there are some little edits,
13:51right, that I want to make. So,
13:52basically what I'll say are things like
13:54this.
13:55Um edit number one will be something
13:56like, "Could you please add some
13:59suspense sound uh throughout this? I'd
14:01like for you to pick a track and add it
14:03to this hook."
14:05Now, I usually like to go through the
14:06entire clip and basically give it
14:09timestamps like, "Oh, I saw this second
14:1022, you did this. I'd like for it to be
14:13like that." It could be anything from
14:15the look to sound to uh kind of the
14:17overall feel or the way it's cutting
14:19things. And again, at the end of all
14:21this, I can basically tell it, "Hey,
14:23could you please uh you know, make this
14:26a little bit better?" And of course, at
14:27the end of this, what's cool is I can
14:28just say, "Hey, can you learn from what
14:30we just did and add it to the skill?"
14:32So, that was a one-shot. What you see at
14:33the beginning of this video is the
14:34actual final product. And before I used
14:37to do this by sections, now I do this. I
14:38just dump the entire footage inside. And
14:41if I've also recorded inside of Tella,
14:44I'm going to tell it, "Hey, here are the
14:46names of the files." Because since it
14:47has direct access, I don't need to
14:49upload the actual documents. Because
14:51yes, there is a human in the loop and
14:53it's me and I actually love it. And
14:54after shooting this right now, I fed
14:56this entire raw footage to Claude code.
14:58And tomorrow, I'm going to wake up and
15:00see exactly what you've seen on screen.
15:02Tomorrow morning, I get to just watch
15:03it, give a couple notes, and get the
15:04final version. So, I'm the director,
15:07Claude's the editor, and I didn't have
15:08to edit any. Which is great because I
15:10actually don't know how to edit at all.
15:11So, at this point, everyone asked me
15:12about the cost. So, let's break it down
15:14to the dollar. First off, I'm on Claude.
15:16I'm on the max plan. It cost me $200 a
15:19month. Now, that's a flat fee, but I
15:20actually use this for all of my work,
15:22not just my video editing. So, what I
15:24did is that I got Claude to actually
15:25measure the number of tokens that it
15:27used, and I have a certain limit of
15:29tokens per month. Now, my last video,
15:31after calculations, cost me $24 to
15:33build, and that's at most. Now, the
15:35great thing is that Hyperframes is
15:37completely for free, so is Parakeet, so
15:39nothing to pay for there. For Tellus,
15:41I'm on the annual plan, it cost me $13
15:43per month, which is roughly $2 per
15:45video. Epidemic Sounds, which is where I
15:47get all of my sound effects and the
15:50music cost me $9.99. Again, I'm on the
15:53annual plan, so let's add an extra
15:55dollar to that. So, all in, I'm pretty
15:57much paying $27 per video. Now, the
16:00reality is that for YouTube editors, a
16:01good editor costs anywhere between 250
16:04and 1,000 plus dollars per video,
16:07meaning that I'm saving hundreds
16:08[snorts] of dollars per video. But, the
16:10huge benefit for me is actually the
16:12speed of execution. I'm now able to
16:14plan, shoot, edit, and release a video
16:17same day, whereas before it took me
16:19anywhere between three and seven days,
16:21depending on which editor I was working
16:24with. And the revisions usually went
16:26back, forth, back, forth, and because of
16:28human error, I actually oftentimes had
16:30to repeat things that I was seeing that
16:32hadn't been changed from one video to
16:34the next. Now, because I don't want you
16:35to suffer through all the same things I
16:37did, I'm going to be giving you this
16:38skill for free inside of my free
16:40community. It's called We Are No Code,
16:42you can find the link below. Now, if you
16:44want to both see the editing skill in
16:46action for a full video, and also get a
16:48full breakdown of what was announced for
16:50OpenAI Dev Day, you can check out this
16:52video right here. Hope you enjoyed this
16:54video, I'll see you guys in the next
16:56one. Let's go.

About this transcript

This page contains the full transcript of I Fully Automated My Video Editing Using Claude Code (Full Walkthrough) by Christian Peverelli, generated from the public captions YouTube serves with the video. The transcript has 3,799 words across 523 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.