13:35Paul J Lipsky · Tutorial
GPT 5.6 Edited This Entire Video
A YouTube tutorial on wiring ChatGPT and Claude into a $79 desktop editor called Borumi so they cut, zoom, and animate your talking-head videos for you.
August 19thThe argument in one line.
Solo creators can reduce video production time by building a Claude AI assistant with custom tools that automates thumbnails, titles, hooks, and editing checklists across a single workflow.
Read if. Skip if.
READ IF YOU ARE…
- A solo creator or small team making YouTube videos 1-4 times per month who wants to cut editing time from hours to minutes using existing tools.
- A founder or marketer shipping video content alongside other responsibilities and looking for a replicable, Claude-powered checklist to standardize titles, hooks, and descriptions.
- Someone comfortable with Claude and Descript who's willing to build a custom AI Project to automate the packaging and editorial side of their production pipeline.
SKIP IF…
- You're a full-time video editor or production team—this workflow is built for solo creators bootstrapping content, not enterprise-scale operations.
- You primarily make narrative, fiction, or heavily scripted content that requires shot-level creative direction rather than talking-head or presentation-based video.
- You've already systematized your own editing and thumbnailing process and are looking for advanced techniques rather than a foundational end-to-end workflow.
The full version, fast.
A solo creator's YouTube workflow becomes manageable when one Claude Project acts as the brain for every video, holding the checklist a full production team would otherwise own. The mechanism is a single assistant loaded with custom instructions, examples, and named tools � a to-do updater, hook writer, title generator, and description writer � so titles, thumbnails, hooks, and descriptions all come from the same context-aware chat. Packaging matters more than production polish, so build thumbnails in Figma with up to six words, brand-jack recognizable logos and color schemes, and pair them with an expressive face shot. Record in OBS, then edit in Descript using Underlord for filler removal, studio sound at 45-50 percent intensity, light compression, a limiter, and auto-generated chapters before exporting straight to YouTube.
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 – 00:21
01 · Hook
Proof/promise/plan cold open. Total pain → found Descript → here's the whole workflow.
00:21 – 01:30
02 · Why packaging matters
MrBeast production guide as the high-intensity benchmark. For part-timers the lever is packaging — titles and thumbnails. 'If nobody knows about it, you've lost the whole game.'
01:30 – 04:00
03 · Thumbnails in Figma
Falling out of love with Canva, into Figma. Six-word max, brand-jack pattern (Microsoft logo + sad anime face = 25K views), color-scheme heuristics, all-in-podcast template clone for show-style content.
04:00 – 06:52
04 · YouTube Content Assistant (Claude Project)
The centerpiece. Custom instructions + project knowledge + named tools (update_todos, write_hook, generate_titles, write_description). The to-do list IS the workflow — it dynamically updates as he chats. He proactively kicks the conversation off because the assistant can't.
06:52 – 08:00
05 · OBS recording setup
Scenes (face cam in corner + desktop), Blue Snowball $40 mic, Canon G7X on Ulanzi rig, desk tripod clamped to monitor, VAI logo overlay.
08:00 – 10:40
06 · Descript Underlord edit pass
Upload OBS file, transcribes, edit for clarity (remove filler), Studio Sound at 45-50% (avoid robot voice), compressor, limiter at -3dB ceiling, add chapters and timestamps, export to YouTube.
10:40 – 10:53
07 · CTA
Generic like + subscribe ask.
Lines worth screenshotting.
- A Claude Project called 'YouTube Content Assistant' acts as the single thread that connects packaging, scripting, editing checklist, and publishing for every video.
- Every YouTube video is fundamentally a checklist — a Claude Project with dynamic to-do updating makes that checklist run itself.
- The proof-promise-plan hook formula should be embedded in the assistant's system prompt so it enforces structure on every script without a separate reminder.
- Title generation, description writing, and hook drafting stored as named tools inside one Claude Project eliminate context-switching between separate tools.
- Descript's text-based editing removes the need to scrub a timeline — you edit the transcript and the video edits itself.
- Thumbnails should be designed before recording, not after — working backwards from the packaging forces clarity on what the video actually promises.
- Six-word maximum on thumbnail text forces the creator to distill the video's core value into a phrase rather than a sentence.
- Brand-jacking a recognizable company's color palette on a thumbnail triggers pattern recognition before the viewer reads a single word.
- OBS screenshot captured mid-session is the most efficient thumbnail face photo because it reuses footage already recorded.
- The YouTube production loop compounds when each tool (Claude, Descript, Figma) handles exactly one phase rather than all phases partially.
- A Claude assistant trained on your personal proof points ensures every video hook opens with credibility specific to you rather than generic claims.
- Solo creators running this stack can match the output cadence of a two-person team because the assistant handles all delegation-ready tasks.
Steal the Claude Project anatomy.
Mod Producer playbook
A Claude Project with rich knowledge, custom instructions, and named pseudo-tools is a no-code team-in-a-box for any repeatable deliverable.
- Build a Claude Project per deliverable type — sales letter, episode, batch shoot, dictation routine — not per video.
- Project knowledge = examples + templates + about-you context. Stuff it.
- Name your tools inside custom instructions even though they're just prompted behaviors. The assistant treats them like functions and the user gets a clear menu.
- Make the to-do list dynamic. The checklist IS the product — emit it as a Unicode-rendered list and update it every turn.
- Have the assistant message FIRST (kick the chat off proactively) because Claude can't.
- Embed your own hook framework as a tool so every script opens the same way.
- Park the brand logo in-frame the entire video. Soft pitch via association beats a hard mid-roll for evergreen tutorials.
Terms worth knowing.
- Descript
- A video and podcast editing application that lets users edit media by editing the transcript text — cutting words from the document removes the corresponding audio and video automatically.
- Claude Project
- A persistent workspace inside Claude that stores custom instructions, uploaded files, and conversation history so the AI maintains consistent context across multiple sessions for a specific use case.
- YouTube packaging
- The combination of a video's title, thumbnail, and description that determines whether a viewer clicks on it — often considered more important than the video content itself for growing a channel.
- Thumbnail (YouTube)
- The still image representing a video in search results and recommendations; click-through rate on the thumbnail is a primary signal YouTube uses to determine how widely to distribute a video.
- Brand jack
- A content strategy where a creator uses a well-known brand's logo or color scheme in their thumbnail or title to borrow that brand's recognition and attract clicks from viewers already interested in it.
- Figma
- A browser-based design tool used for creating user interfaces, graphics, and layouts — commonly used by creators for thumbnail design because of its speed and collaborative features.
- Canva
- A drag-and-drop graphic design platform popular with content creators for making thumbnails, social media graphics, and presentations without professional design skills.
- Click-through rate (CTR)
- The percentage of people who see a video thumbnail or link and click on it, a key metric indicating how compelling the packaging is relative to its distribution.
- AI slop
- Informal term for large volumes of low-quality, generic content generated by AI with minimal human judgment, increasingly saturating the internet and eroding trust in online information.
Things they pointed at.
00:25toolMrBeast YouTube production guide
07:05productBlue Snowball mic
07:40productCanon G7X
07:44productUlanzi rig mount
07:52channelVAI community
Lines you could clip.
01:18
“Your product or service could be better than everybody else's, but if nobody knows about it, then you've still lost the whole game.”
00:06
“When I first started, it was a total pain. Took me half a day just to make something.”
05:20
“Every single YouTube video is essentially a checklist. Checklists operate the whole world if you think about it.”
09:56
“If you do it in their default where it's at a hundred percent, then you're gonna sound like a robot.”
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.
metaphor
A lot of people ask me how do I make YouTube videos? When I first started, it was a total pain in the Took me half a day just to make something that got half of you. And so after that, I realized, wow.
I should be using a tool where I can just type in what edits I want and I found that tool. It's called Descript. I wanna jump into that.
I wanna jump into the whole process of how I make the titles. How I use Claude the whole way through and how it just makes it a whole lot easier. It's opened up a lot of doors for me.
I think that building a personal brand and having your message put out there is really important now that we've got a bunch of AI slop. Let's jump into it. So if you go and you read the mister beast production guide on making YouTube videos, his is gonna be like way on the right side of the spectrum in terms of intensity and a lot of us are not full time YouTubers.
We just do this on the side to market a product or get our ideas out there, build a network. But I will say the one part that's been really helpful for me and I'm trying to get better at is the packaging. Because your product or service could be better than everybody else's, but if nobody knows about it, then you've still lost the whole game.
So it matters a whole lot and it comes down to the titles and the thumbnails. So I'm gonna go through titles in a second, but let's get into the thumbnail first. You're supposed to actually do this before you even record the video, but I do have a YouTube assistant where if I just have an idea and I wanna get it out, then I'm working backwards into the packaging.
But it is a little bit worse, I feel like, that way. So let's jump into the thumbnails just real quick. So I've experimented with a couple different ways of doing this, and it starts with just doing some research on what are the ones that are working.
And I've used Canva and I've used Figma. I've also used Photoshop. Photoshop takes too long.
Canvas the quickest and I'm now really falling towards Figma because I just am fastest in there but it was missing this remove background feature that Canva had. Now that they've added it, I'm just trying to go all in on Figma. But what I found to work is having some simple text up to six words.
If you're doing anything with logos that are associated with a brand, then you can do like a brand jack. So one of the best videos that I've had, higher performers for my small channel, have 25,000 views in a couple weeks.
I and it was about Microsoft. Microsoft drops prompt management. My channel's about AI stuff, put a big logo of Microsoft, and then something that shows emotion.
So it's like, here's this anime guy, and he's sad. So I've followed that to be one of the best ways to do it. When I get like super fancy or I try to be super fancy and I have these like weird looking fonts, it's just overwhelming and people get annoyed and they don't click on it.
The new thing that I'm trying out is just doing this which is if I'm talking about a brand that everybody knows then I'll use the color pattern that's associated because there's certain heuristics. There's patterns in people's brains where they'll just see it.
They associate with the product. If they like the product that they wanna learn more about it then they'll click on it. So in my case, cool logo plus color schematic.
We'll see how that works. And the same thing goes even with we have a weekly show that we do inside of our community called VAI, and there's a popular show called the all in podcast.
So I'm like, oh, cool. Let's just copy that sort of template for this. And then eventually, for each channel or each type of video that you're creating, you'd have a different project or different page for each type of thumbnail.
So as I make more of these, then I'll just be able to come in here and I can just have a template, command d, create a new one. That's good to go. I'll typically grab a thumbnail of my face on the subject.
Right? So what I'll do towards the end of the video is I'll just switch into this mode and then I can right click on OBS. You can't see it but the thing that I'm recording in and then just hit capture screenshot.
The lighting's not great right now. Lighting also matters. But then you can just touch it up, remove the background, crank up the exposure a little bit.
If you wanna look like your face is melted, which is what a lot of people do, then you can do it. I think it's just enhanced resolution. There's some effects in here.
Now that thing that you saw at the beginning, the assistant also, when you wear a hat forward, you're basically a creative director. So I'm a creative director now. No.
I'm not. But the thing you saw at the beginning, this is . This is video content ops.
It is a cloud assistant and it has a couple secret sauce elements to it. All these are available in the description. You can just copy pasta them.
But what it is it has project knowledge and it has custom instructions. So in nerd world, nerd ville, you can go and write your own assistant but we wanna do no code here because we don't need to. So what is this?
This has a bunch of stuff in it. This is basically a giant prompt with tools in it. I have a tool that allows you to make the screen bigger now.
I have a tool in here that updates the to dos for a YouTube video. Pretty cool. Right?
Because every single YouTube video is essentially a checklist. if you think about it. And in this checklist you have the things that normally a team would do.
So if you looked at a big YouTuber then they'd have a team and they'd delegate these different tasks. I can have it update them as we go. So when I open up the assistant and I either have a video idea or I have a title or I have a script, it will know what to do and then update the to dos accordingly and help me get through all of them.
So it's dynamic. It is trained to know that, hey, I'm gonna be the first one that makes the message, the user, because it doesn't have the ability to proactively message.
And it will have that one tool of the to dos. It has examples of it. It uses Unicode which is essentially like to show you what's going on.
So the different tools have an f for function. If you're a nerd, know what a function is. And then it will update them as it goes.
It has some things in here around how to write hooks. Hooks matter a lot. What's the proof, the promise and the plan that you're gonna do at the beginning of the video to get people to go and actually grab their attention.
So why should I listen to this person? Because there's a bunch of people out there that are shysters. So why should I listen to Parker?
So you basically say what you've done that's interesting and then it will ingrain that hook of the proof, the promise, the plan into every video that it replies with.
Then you have examples of titles for the title generator tool. So title generator only, you can update the to dos and then doing what else is in here?
Descriptions, you have the hook examples, you have the rules around how to select which tool you wanna use, some best practices, all this.
So, yeah, that'll be in the description. You can edit it as you see fit. The only piece that you'll need to edit is the context about you because I don't know who you are and what you've done.
And then other things that are in here that make a little bit more saucy is when you wanna get examples of titles that fit your niche, then go and find them. So look around, find the people that are killing it in your niche, and then paste them all in here. So those are descriptions.
Whoops. You can do descriptions as well. And then I have titles and then I have a how to of how to use this assistant.
So that's in the description as well. And it also helps the assistant work.
So this is really nice and go and set it up for yourself.
I record in OBS and let me actually, I need to change a setting so that you can see it.
Cool. Now you can see it. And the way that it works is I have a few different scenes.
So on this scene I have scene b, which is me right here in this corner.
And then I have the desktop. I have the microphone, which is a blue snowball, think you call it.
It's like a $40 mic. It's fine. And then I'll touch it up in the software that I'm gonna show you afterwards.
I have a Canon g seven x. I can't show you that. It's on a u u rig mount.
It's basically just a cheap mount but it works from Amazon and then a desk tripod that clamps to my desk right behind the monitor. And then I have a logo for VAI, the best community on the planet.
So when you get done with the video, I use Descript right now because they've released a lot of new products. Everyone's asking for what's the cursor for video editing and this is it. They have what's called the underlord.
And when I go into here, I basically just click new project. This button would say new project. I upload the file.
So when I finish recording, it dumps into this raw OBS footage thing here.
And so I grab that. The files being written right now. That's this.
That's me talking. I've messed around a lot with the frame rate. Right now it's at 30.
It's still four k but instead of being like six gigs for a half hour video, it's five eighty. So bringing down the quality where it's always this comparison of quality versus file size. But I find that to be something you can play around with.
Get a nice file size so you can get these done quicker. And then when you drop it in, it transcribes it. Cool.
And there's a couple settings that I'll go through. But in this case, twenty eight minute video.
I'll come in and I can hit edit for clarity. And what that does is it cuts to the chase, removes filler words, digressions, blather apparently.
You can mess with the intensity. I usually just use remove filler words but this will run, it'll show me where in here. So woah, big video coming out.
There's some of this stuff where it's I actually want to have it in there. I might remove this.
This would not make sense to remove. So you just read through these. That's why I don't actually use that one very much.
Instead, I'll go in and I'll listen to it. The channel, my name is Parker Rex. I'll add tech for a startup that's over $23,000,000.
And after that, I'll add studio sound. But if you do it in their default, where it's at a 100%, then you're gonna robot.
So you don't wanna sound like a robot. So I'll crank it down to 45, maybe 50 on the intensity.
And then for audio, you have a couple settings. Again, you can program all of these. That's the plan.
I'll do a compressor once this studio sound thing is done. If you're in the audio world, it's just a nice thing to make it a little squishier, but you can overdo it. So I'll just listen to it, you know, coming out.
What's he saying there? We're not sure. Cool.
And then I'll do a limiter. A limiter limits the bars, basically.
Oh, I want it to just be stuck at a three d b ceiling. Cool. The last bit I'll do is I'll just jump back into Underlord.
I'll grab add chapters. I'll submit it and it'll give me a bunch of chapters and then I say add timestamps and then this is what goes into the export. After that, I'll just click here and then I'll select YouTube.
I'll paste this in here. If you guys found this helpful, make sure you like the video. Make sure you subscribe to the channel so you can stay up to date and I'll see you in the next one.
Peace.
The Hook
The bait, then the rug-pull.
Parker opens with the problem (half a day to make one video), the discovery (a tool you can talk to in plain English), and the promise (the whole packaging-to-publish loop in one Claude Project). Proof / promise / plan — the same hook structure he later codifies as a tool inside his assistant.
Named ideas worth stealing.
05:40model
Proof / Promise / Plan hook
- Proof — why should anyone listen to you
- Promise — what they'll get from the video
- Plan — the steps you'll walk through
Parker's three-part opener. He builds it into the assistant so every script he writes opens with the same skeleton.
Steal forevery script opener — embed it as a named tool inside a Claude Project
04:00model
YouTube Content Assistant (Claude Project anatomy)
- Project knowledge: title examples + thumbnail examples + description templates + about-you context
- Custom instructions: when to fire which tool, best practices, hook examples
- Named tools (prompted, not API): update_todos, write_hook, generate_titles, write_description
- Dynamic to-do checklist that mutates as the chat progresses
- Assistant kicks the chat off proactively because it can't message-first on its own
A no-code, no-API team-in-a-box built entirely with a Claude Project. The to-do list functions as the team a big YouTuber would have. Tools are just prompted behaviors, not real function-calling.
Steal forany repeatable creative deliverable — sales letter, episode, batch shoot, dictation routine. Replace video with whatever Joe ships repeatedly.
03:20concept
Brand-jack thumbnail rule
Pair a recognizable brand logo with an emotion overlay (anime crying face, melted-face effect). Hijacks the existing neural shortcut viewers have to that brand.
Steal forthumbnails for topical / reactive content
02:55concept
Six-word maximum
Thumbnail text caps at six words. More than that, weird fonts, fancy effects = overwhelm = no click.
Steal forevery thumbnail Joe ships
09:20concept
Studio Sound at 45-50%
Descript's Studio Sound default (100%) makes you sound like a robot. Crank to 45-50% intensity for the sweet spot.
Steal forany Descript audio cleanup pass
How they asked for the click.
VERBAL ASK
10:40subscribe
“If you guys found this helpful, make sure you like the video. Make sure you subscribe to the channel so you can stay up to date and I'll see you in the next one. Peace.”
minimal. No product pitch despite the VAI logo being in-frame the entire video and being verbally name-checked mid-walkthrough as the best community on the planet. Soft pitch via association only.
MENTIONED ON CAMERA
Visual structure at a glance.
