11:05Duncan Rogoff | Learn Claude Code · Tutorial
Claude Code + HeyGen = Content Creation CHEAT CODE!
How to build a Claude Code skill that researches, scripts, and renders a yap-style AI avatar video from a single slash command.
June 22ndThe argument in one line.
Claude Opus 5.5 can now handle every stage of short-form video production, from script splitting to motion graphics to captioning, cheaply enough for one creator to replace a paid video editor entirely.
Read if. Skip if.
READ IF YOU ARE…
- A YouTube or Instagram creator who currently pays a human editor for short-form cutdowns and wants to see whether an AI pipeline can replace that cost.
- Someone comfortable directing Claude Code through a multi-step build across several tools and APIs, not looking for a plug-and-play consumer app.
- A creator running a paid community or funnel who wants shorts that feed a specific downstream offer, not just raw reach.
SKIP IF…
- You want a ready-made app you can buy; this is a custom-built, self-maintained system stitched together from several separate services.
- You're not willing to track per-video AI costs across five or six different tools to know what a short actually costs you.
The full version, fast.
Duncan Rogoff argues Claude Opus 5.5 finally makes AI-built short-form video good enough to drop his $1,500-a-month editing budget. His system takes a YouTube video, GitHub repo, or article, grabs the strongest hook, writes three scripts, splits each sentence into individual visual moments, and renders each one with a HeyGen AI avatar, open-source motion graphics, sound effects, and AI-generated music, then hands scheduling and comment-triggered DMs to Blotato. He splits the work across Claude sub-agents (research, build, render, check) so one main session can plan and quality-check everything the others produce. The full pipeline costs about $8.34 per short on a Claude Max subscription versus roughly $21 pay-as-you-go and $100 for a human editor, meaning 30 shorts a month costs $250 instead of $1,500 for 15 edited videos.
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 – 00:56
01 · Claude Opus 5.5 shorts results
Duncan opens with a blunt claim: Opus 5.5 is the best model he's used, beating Fable 5.1 and GPT-6 Astra, and in 48 hours it let him build videos he'd spent months failing to make.
00:56 – 01:56
02 · AI-generated short example
He plays the actual AI-built short, a '342-hour AI course' Instagram Reel, so the viewer sees the finished avatar, captions, and motion graphics before any explanation starts.
01:56 – 02:55
03 · Editor costs vs. AI results
He states his old cost structure, $100 per video and 15 videos a month, then shows the new AI video's early stats: 2,500 views, 80 likes, 85 comments, 119 saves, in four hours.
02:55 – 04:11
04 · Shorts-to-YouTube funnel strategy
He explains the business reason for the system: Instagram grows reach but rarely converts, so shorts exist to drive viewers to his YouTube channel, which funnels into his $9-a-month Cloud Code Club.
04:11 – 04:47
05 · How to research winning creators
Before building anything, he studies two accounts, Nick Saraev at 685K and Kallaway at 134K, pulling apart both their visual style and how they script.
04:47 – 05:33
06 · The 8-step shorts system
He lays out the full pipeline: grab the best moments, find working hooks, write three scripts, generate the AI avatar, split every sentence, draw each moment, render, then watch and fix.
05:33 – 06:31
07 · How to structure short scripts
He breaks a script into named parts, hook, lock-in proof, head fake, re-hook, list, payoff, and CTA, using his own '342-hour AI course' script as the worked example.
06:31 – 07:18
08 · Splitting scripts into visual moments
Each sentence gets split into separate visual moments, because one sentence can carry more than one idea, and every moment gets its own graphic.
07:18 – 08:03
09 · FFmpeg editing and HeyGen avatar
All cutting, trimming, and stitching runs free locally through FFmpeg; the talking-head cutout uses Apple's free Vision framework; the AI avatar itself comes from HeyGen, trained on 15 seconds of his own footage.
08:03 – 09:03
10 · Refining motion graphics and captions
He shows the before and after: a static screenshot becomes a number that counts up and gets highlighted, and captions go from plain text to animated, italicized word-by-word reveals, built with the open-source Hyperframes toolkit.
09:03 – 09:49
11 · Sound effects and AI music
Small sound-design layers, whoosh, click, shutter, pop, get added under the voice, and the background music bed is generated with an AI model called Suno through Kie.ai for a fraction of a cent.
09:49 – 10:37
12 · Editing rules that boost retention
Six standing rules: the hook never cuts, proof comes before explanation, cuts only happen when the idea changes, one idea per picture, alternating light and dark graphic modes, and big title cards for the moments that matter.
10:37 – 11:36
13 · Splitting work into sub-agents
One main Claude session handles planning and quality-checking while separate sub-agents own research, topic sourcing, script writing, avatar rendering, and final review, so the pieces build in parallel.
11:36 – 12:21
14 · Thumbnails, captions, and auto-scheduling
GPT Image 2.5 generates on-brand cover thumbnails, platform-specific captions get written for Instagram, TikTok, X, YouTube Shorts, and LinkedIn, and Blotato handles both scheduling and automated DMs triggered by a keyword comment.
12:21 – 13:56
15 · Full cost breakdown per short
He itemizes every cost, avatar render $4.84, cover $0.14, research $0.12, Claude usage $13.30, scheduling $3.23, music under a cent, for a Claude Max subscription total of $8.34 per short versus $21 pay-as-you-go and the $100 he used to pay an editor.
Lines worth screenshotting.
- Duncan Rogoff says Claude Opus 5.5 replaced the $1,500-a-month human editor he was paying to make his Instagram shorts.
- His full AI-avatar-plus-motion-graphics short costs $8.34 to produce on a Claude Max subscription, versus $100 for a human editor.
- Without a Claude subscription, the same short would cost about $21 in raw API spend, still a fifth of the editor price.
- The single most expensive line item in the pipeline is the AI avatar render at $4.84 per short, run at 1080p through HeyGen.
- Rendering the avatar at 1K resolution instead of 2K would cut that cost roughly in half.
- He splits every script sentence into distinct visual moments because one sentence can carry more than one idea, and each idea gets its own graphic.
- His editing rule is that the hook never cuts until the full opening sentence finishes, so the viewer understands the topic before the first edit.
- He alternates between a dark-mode and a light-mode graphic style across the video specifically to keep the viewer visually engaged, a tactic he credits to Nick Saraev.
- He studies competitors like Nick Saraev and Kallaway not just for visual style but for how they structure and script their videos.
- His shorts strategy exists to feed his YouTube channel rather than convert directly, because he estimates it takes roughly seven hours of viewer attention before someone buys.
- All of the video editing, trimming silence and stitching clips, runs for free locally through FFmpeg instead of a paid editing tool.
- The talking-head cutout effect uses Apple's free, built-in Vision framework instead of a paid green-screen or rotoscoping service.
- He generates the background music with an AI model called Suno through the Kie.ai platform for under a cent per short.
- One Claude main session plans and quality-checks the output of several sub-agents that each handle research, scripting, avatar generation, and rendering.
- Captions and platform-specific copy are auto-written for Instagram, TikTok, X, YouTube Shorts, and even LinkedIn from the same source script.
- Comment-triggered DMs, run through a platform called Blotato, automatically send a resource link to anyone who comments a specific keyword.
Eight Steps Turn Claude Into A Video Editor
THE AI PIPELINE
A single AI-avatar short costs about $8 to produce end to end, and getting there means treating scriptwriting, motion graphics, sound, and scheduling as discrete steps a model can execute one at a time.
01Claude Opus 5.5 shorts results
- A single capability jump in the underlying AI model can turn months of failed attempts into a working system almost overnight.
- The claim being tested is specific and falsifiable: matching a $100-per-video human editor, not just making AI videos in the abstract.
02AI-generated short example
- Showing the finished output before explaining the process lets a viewer judge quality on their own terms instead of taking the creator's word for it.
- A single Instagram Reel can combine a cloned voice avatar, animated captions, motion graphics, sound effects, and music, all generated rather than filmed.
03Editor costs vs. AI results
- A recurring editor cost of $100 per video adds up to $1,500 a month at 15 videos, a number worth totaling up before comparing it to any automation.
- Early engagement numbers, views, likes, comments, saves, shares, are the real test of whether an AI-made video performs, not just whether it looks polished.
04Shorts-to-YouTube funnel strategy
- A platform that grows reach isn't necessarily the platform that converts buyers; one can build audience while another does the actual selling.
- Viewers reportedly need around seven hours of cumulative attention before they buy, which reframes short-form content as a funnel step rather than the sale itself.
05How to research winning creators
- Studying competitors' scripts, not just their visuals, surfaces structural patterns that a purely visual copy would miss.
- Two creators can use completely different visual styles while sharing the same underlying script structure, which is the part worth extracting.
06The 8-step shorts system
- Breaking a complex production process into discrete steps, each small enough for an AI model to execute reliably, is what makes the whole pipeline repeatable.
- Ending a pipeline with a dedicated review step catches errors that would otherwise ship silently.
07How to structure short scripts
- A short-form script has named, reusable parts, hook, lock-in proof, head fake, re-hook, list, payoff, CTA, that can be filled in for any topic.
- A mid-script re-hook moment exists specifically to catch viewers right before they'd naturally swipe away.
08Splitting scripts into visual moments
- A single sentence can contain more than one visual idea, and each idea deserves its own moment rather than being crammed onto one screen.
- Treating one idea per picture as a hard constraint forces clearer decisions about what actually needs to be shown.
09FFmpeg editing and HeyGen avatar
- Free, decades-old tools like FFmpeg can still do professional-grade video trimming and stitching when directed correctly, no paid editing software required.
- An AI avatar convincing enough to replace a paid editor can be trained from as little as 15 seconds of a person's own footage.
10Refining motion graphics and captions
- A static screenshot becomes noticeably more engaging when the specific number in it animates and gets visually highlighted instead of just sitting on screen.
- Captions that animate word by word and use italics for emphasis read as meaningfully more polished than flat, all-at-once text blocks.
11Sound effects and AI music
- Small, almost subliminal sound effects, a whoosh on a zoom, a click on a transition, can measurably help retention even when most viewers won't consciously notice them.
- AI-generated music beds can now be produced cheaply enough, under a cent per video, that music is no longer a real cost constraint on short-form content.
12Editing rules that boost retention
- Never cutting away until the opening sentence finishes lets the viewer register what the video is about before the visual style takes over.
- Leading with proof of value before explaining the mechanism keeps skeptical viewers watching long enough to reach the explanation.
13Splitting work into sub-agents
- Splitting a complex build across multiple narrow-focused AI sub-agents, each owning one job, lets pieces get produced in parallel instead of one long sequential process.
- Keeping one main session responsible for planning and quality-checking gives a single point of oversight over everything the sub-agents produce.
14Thumbnails, captions, and auto-scheduling
- Platform-specific captions for Instagram, TikTok, X, YouTube Shorts, and LinkedIn can all be generated from one source script rather than rewritten by hand for each platform.
- Automated DMs triggered by a specific comment keyword turn a passive comment into an active lead-capture moment without manual follow-up.
15Full cost breakdown per short
- Itemizing every cost in a pipeline exposes which single line item actually drives the total, here an avatar render at $4.84 out of $8.34.
- A flat monthly AI subscription can be cheaper in practice than metered API pricing once usage is heavy enough, $8.34 per short on a subscription versus $21 pay-as-you-go.
Terms worth knowing.
- Shortify
- The creator's own name for the Claude-based pipeline that converts a YouTube video, GitHub repo, or article into a fully produced short-form video.
- HeyGen
- An AI avatar platform used here to generate a digital clone of the creator's voice and face from about 15 seconds of source footage.
- Hyperframes / FRAME.MD
- A free, open-source motion graphics toolkit, made by the HeyGen team, used to animate on-screen text and design elements.
- Blotato
- A third-party scheduling and automation platform used to publish videos across social platforms and trigger DM automations when someone comments a keyword.
- Apple Vision
- Apple's built-in, free on-device framework used here to cut a talking-head subject out from its background without a paid tool.
- Sub-agent
- A separate Claude Code session assigned one narrow job, such as research, scripting, rendering, or quality checking, that a main session coordinates and reviews.
Things they pointed at.
00:24toolAI Engineering From Scratch (free GitHub repo)
04:16channelNick Saraev
04:21channelKallaway
08:01toolHeyGen
08:59toolHyperframes / FRAME.MD
09:38toolSuno via Kie.ai
12:06toolGPT Image 2.5
Lines you could clip.
00:00
“Claude Opus 5.5 is the best model on the planet and it's not even close.”
02:55
“I pay about $100 a video. I get 15 videos a month. So I spent about $1,500 to create a video every two days.”
12:38
“So for me, one short with my subscription costs $8.34. Without the Cloud subscription, it'd be about $21.”
09:49
“The hook never cuts.”
10:37
“Split the work. Run it at once.”
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.
analogystory
Claude Opus 5 .5 is the best model on the planet and it's not even close. I literally spent months trying to build videos that look this good and I could not do it. Not with Fable 5 .1, not with GPT -6 Astra, but in the last 48 hours I've been working with Opus 5 .5 and it has been unbelievable.
So today I just want to walk you through all of the pieces of the system that I put together to build short form videos like this one with the AI avatar, with all the graphics, with music, with everything so that now I can run my Instagram completely on autopilot. And at the end, we'll break down all the costs and everything so you don't have to worry.
This free GitHub repo is a 342 -hour AI course. 56 ,000 people have already started. It's 523 lessons in 20 phases from the math all the way up to agents and MCP servers.
So you'd think you'd have to start at lesson one and grind through all 523. You don't. Pick one goal, like build agents, and it gives you the path.
But here's the part that got me. Every lesson makes something you keep. A prompt, a skill, an agent.
And it turns clawed code into your tutor. One install, then type start learning. 342 hours of lessons and it costs...
nothing. If you want the link to the repo, comment scratch. So I'm blown away by the results.
And there's just a couple of things that you should notice when you watch this, right? Of course, how all of the graphics animate. We have a title on top here, my AI avatar, how it's cut out from the background, all of the captions that are going on.
There's music, there's sound effects. We use screen capture and screenshots. Like it's actually insane what this is making.
I did put together a free guide for building a system like this. It has all sorts of prompts and stuff to get you started. If you want to get access to it, I'll leave a link in the description.
I spent 15 years as an art director and motion graphics designer at companies like Apple and PlayStation. So I have really high standards for what good quality video looks like. And literally up until Opus 5 .5, I was not happy with it.
any of it. But now with this, I am proud to put this work out there. It looks so good.
So this is my Instagram channel. I have about 30 ,000 followers and I actually get asked all the time, how do I make these videos that look like this? And really the secret is I've actually been paying editors to make these videos, but now I don't have to.
So just to be transparent, I pay about $100 a video. I get 15 videos a month. So I spent about $1 ,500 to create a video every two days.
But Literally today, I published my first AI generated video, this one right here that we just saw. And these are the results.
I published this four hours ago. We're already at about 2 ,500 views and the engagement is pretty insane. 80 likes, 85 comments, 119 saves and 25 shares.
So this is performing as well or outperforming a lot of the content that I was paying people to make. And so, like I said, I had been trying to build something like this for months, and these were the results that I was getting before, and they weren't nearly as good. And I tried posting a few, and the engagement was horrible.
So I stopped. So we're just going to break down all the steps today to getting something that actually performs well. So let me talk about why this all started and like how I actually plan to make money from all of this.
So the whole reason I wanted to make this skill at first was basically just to take my long form YouTube videos where you're watching this right now and turn it into shorts. And there's a reason for this. So if you don't know, I run a community called the Cloud Code Club.
It's nine dollars a month. And Instagram is great for me for growing my reach. But it doesn't actually convert to buyers.
Most of the people who convert to somebody who. actually pays me come from my YouTube channel. So the whole thinking behind this is I want to be able to create Instagram shorts that then send people to my YouTube channel because it takes on average, like people spending seven hours with you to buy something.
And so like how many 60 second shorts do they need to watch before they've spent seven hours with you versus how many long form YouTube videos. So people come to my YouTube channel, they watch a video like this one where I used Opus 5 .5 to build an AIOS and they click a link to come to the call. odd code club and they sign up.
That's sort of the flow and why this whole system was created in the first place. Since then, I've expanded it to work not just with YouTube videos, but with any GitHub repos, websites, articles, products, like legitimately anything. Like it's pretty insane.
So the first thing that we do is a step that most people skip when they want to build systems like this for themselves. And it's research. Like it is the most important part.
Like you want to figure out what's already working. So I just like studied some big creators like Nick Sarayev. If you're watching me right now, you probably know.
who he is. You may not know Callaway. He's great at marketing.
And if you come to his channel, if we can see Nick's channel, like his videos function largely similarly to what I just showed you with mine, right? He has a different style, a different vibe, but a lot of the same concepts apply here. Same thing if we go into Callaway, totally different style, but his videos are great, right?
Like I love these graphics. You can see him down the bottom with his little card cut out and all of these things. And I think there are a couple things that you want to study, not just the design and the look, but also how they script.
things. So these are the eight steps that the system goes through. And I think you will find this theme of like breaking things down into smaller and smaller pieces that these models can really understand and dissect and then recreate.
So it's going to grab the best moments from a video or a repo. It's going to figure out which hooks are working right now, because I think we all know how important hooks are. And so you basically want to take those word for word and use them and then stack the rest of the video on the end.
It's going to write a couple of scripts for me. It's going to have my AI avatar send this. I'll show you how I make that in a second.
It's then actually going to split up every sentence, draw each moment, render the full short, and then it's actually going to watch it and check its own work before giving me the final output. So you can start to see how many different pieces there are of a system like this, but it all starts with the script. So the first thing you want to do is to actually break the script down into distinct.
parts. So first we start with the hook. This free GitHub repo is a 342 hour AI course.
People love numbers. Next is this lock in moment, right? This is sort of like the proof, like why should somebody pay attention to this?
56 ,000 people have already started. If other people like it, you're going to like it too. And then this head fake where you're sort of kind of breaking through a common belief.
So you'd think you'd have to start at lesson one and grind through all 523, but you don't. And now somebody leans in. And then you re -hook them.
But here's the best part. So this is the moment right here where somebody might swipe away, but this little transition sentence keeps them watching. And then a list.
People love lists. Every lesson makes something you keep. A prompt, a skill, an agent.
And then the payoff. It turns Cloud Code into your tutor. 342 hours of lessons, and it costs nothing.
And then, of course, at the end, CTA. If you want the link, comment this word. And that's the part that we wire up at the end.
So now once we have the script, how do we actually... get into the graphics you actually need to have opus 5 .5 analyze the sentence and split it into distinct moments start at lesson one grind through all 523. this moment you Don't write that impact moment.
Pick one goal like build agents and it gives you the path. So you can see that a sentence may actually have multiple moments. And so we actually need to break this out.
And now we start creating these graphics for the different moments and we can kind of transition between the two. Right. Like you think you'd have to start at less than 21 and grind through all 523.
We have this transition moment. Right. And we just hard cut to my face.
You don't pick one goal like build AI agents and it gives you the path. So each little. section gets its own graphic.
And so all of the editing is actually being done for free locally on my machine with a service called FFmpeg. You can literally tell Claude, hey, install FFmpeg. It has been around for decades.
I don't know, but it's free and it is awesome. It trims out all the silences. It zooms in, it zooms out, it stitches the pieces together.
And then for this cutout, we're actually also using a free service called Apple Vision, which I didn't even know existed. I basically just said like, hey, how do I make a cutout like this? And it told me and then it did it.
I'm not going to cover in detail creating my AI avatar. There are tons of videos online about that, but I use a site called HeyGen to do this. I literally uploaded like 15 seconds of me talking to the camera.
I actually just used an intro from one of my other YouTube videos, and it created like a very believable avatar for me. This piece of creating the avatar is by far the most expensive piece of the whole system. So once you start making this, you start thinking about how you can refine the system.
And again, that's the part that I think people miss is this idea of that. It doesn't usually just work the first time out of the box. You have to look at how can you make it better?
Like we started with just like a static screenshot that would slowly kind of zoom in, which was fine, but kind of boring. But now we actually punch in hard to the number. We highlight it.
The actual number counts up right in the star like highlights. And so it just adds this visual interest instead of a list on a page. Again, it's not just a screenshot.
We zoom in. we highlight the individual elements, right? So how can we actually plus this up to make this more impactful?
And how can the system decide between showing a screenshot or showing motion graphics? Other little details to pay attention to, like I upgraded the captions. They used to look like this.
They were boring. And now they have italics and they animate on a word at a time. And this is much more engaging to watch.
So for all of these motion graphics, we are using a free open source repo called Hyperframes. It's by the HeyGen team who does my AI avatar. One other detail that you may not notice but help a lot for engagement are these little sound effects like a whoosh or a camera shutter or a little click or a digital noise or whatever, right?
So take a listen, you can see what I mean. grind through all 523, you don't. Pick one goal, like build agents, and it gives you the path.
But here's the part that got me. Every lesson. So you can hear when we zoom in on me, there's like a whoosh sound.
And when the paths are building on, there's a couple of different clicks, right? And so it's very subtle, but again, it like keeps the viewer watching. And then I used an AI music generator called Sunu through this platform, Kai .ai to generate the music.
It's like insanely cheap to generate, which I'll break down in a second. And then there are actually a couple of rules to think about that make things easier for the viewer to understand, because you always want to focus on like comprehension and making sure that the person who's watching like knows what the heck you're talking about.
So the first thing is that the hook never cuts. We can animate on the graphic, but we actually wait till I get through speaking that opening sentence before our first cut. So the viewer can actually understand what the video is about.
First, we're going to give proof that the thing is valuable before going into any sort of explanation. Anytime the idea changes is when there's a cut and we actually kind of mix and match between this bottom crop and like me full frame in the screen. So you actually get into building in variety into these videos.
One idea per picture. Do not put too much on screen here because again, that confuses and overwhelms the viewer and they won't understand what you're talking about. Then the other thing that I did that I actually got from Nick is this idea that some of the video is kind of like dark mode.
Some of the video is light mode. So it's switching back and forth between these two graphic styles, which again, keeps the viewer engaged. So one of the biggest tips for building a system like this to make things more efficient and actually cheaper for you to operate is actually splitting the work into sub agents.
So I had my main session do all of my planning and then I focused on individual pieces. So the first part is how we are doing research, right? So I can split it into researching other accounts, like all of their topics, like their scripts, like what lead magnets are there?
delivering, right? So we're building up our library of things we can talk about and resources we can share for free based off of what is already working. So then you can split the build out.
So we had a subagent building a topic engine that will actually go get new topics daily. You can set this to weekly or however long you want. It actually researched my Cloud Code Club community for any of the lessons in there.
Now we are turning what we teach into little short form videos to drive traffic to the club. We use a subagent to build out any of our free resources, then other subagents to write the script, create the digital avatar, check the final output, render the music, all of these different little pieces to come together. The system also uses GPT image 2 .5 to generate these nice thumbnails.
So if we have a new tool launch, it's actually going to create a thumbnail like the one we saw. If we don't, if it's just coming from an idea or one of the lessons, it'll just use that opening graphic, which should make it stand out, but still have a lot of visual interest. And you can also see that everything is.
Still in brand and all feels like me, right? My orange hoodie, like this green neon glow, like it should be familiar. Next, the system writes captions for every single platform, for Instagram, for TikTok, for X, for YouTube shorts, literally everywhere, even for LinkedIn.
They all get scheduled through this third -party platform called Blotato, which has my favorite MCP for making sure my content gets published everywhere. And then I can actually use Blotato to create these comment automations. So when people comment a word, it actually sends them the thing.
So let's talk about the cost. The AI twin render is by far the most expensive piece at $4 .84. So the cover image costs about 14 cents.
I am rendering this at 2K resolution. You could render it at 1K for half the price. I'm spending 12 cents to do a little bit of topic research.
In terms of quad API usage, this is also expensive. It's $13 .30 for a script, a plan, an edit, creating all the graphics, checking all the captions, all of this stuff. But really...
I just pay for the Cloud Max plan. I am personally on the 20X plan, but in the end, it becomes way cheaper to do that than spend this much on the API. Scheduling and DMs cost about $3 and music is literally less than a penny.
It's insanely cheap. So for me, one short with my subscription costs $8 .34. Without the Cloud subscription, it'd be about $21.
But if I'm publishing one reel a day, it's about $250 a month to create 30 reels. And if you remember, I was paying my editor. $100 per video.
So $1 ,500 a month for only 15 videos. Whereas with this system, I can publish 30 videos a month for only $250. So you can see it's a fraction of the cost.
If I want to publish three a day, it's about 550 bucks once I want to crank up the volume. And of course, what's so powerful is how scalable a system like this truly is. So if you do want to learn how to use quad code to build websites, apps, games, and more, and get access to all my resources already built for you, just check the link in the description.
If you want to see how I use quad to build some insane Instagram carousels, check out this video right here. I'll see you over there.
The Hook
The bait, then the rug-pull.
Duncan Rogoff opens by declaring Claude Opus 5.5 the best model he's used, then spends thirteen minutes proving it with the actual eight-step system he built around it: an AI avatar, motion graphics, captions, and music that together replaced the $1,500-a-month editor who used to make his Instagram shorts.
Named ideas worth stealing.
04:47list
The 8-Step Shorts System
- Grab the best moments
- Find hooks that work now
- Write three scripts
- The AI twin says it
- Split every sentence
- Draw each moment
- Render the short
- Watch it, fix it, repeat
The full production sequence he runs a video, repo, or article through to end up with a finished, edited short.
Steal forany repeatable short-form video production pipeline
05:33model
Script Structure: Hook to CTA
- Hook
- Lock-in proof
- Head fake
- Re-hook
- List
- Payoff
- CTA
A named sequence of script beats built to state a claim, prove it's worth attention, break a common assumption, catch the viewer before they swipe, deliver a payoff, and close with an ask.
Steal forany short-form video script
09:49list
Six Editing Rules
- The hook never cuts
- Proof, then explain
- Cut when the idea changes
- One idea per picture
- Half light, half dark
- Big title cards for the moments that matter
The standing constraints he applies to every edit to keep the video easy to follow and visually varied.
Steal forany explainer or educational short-form video
How they asked for the click.
VERBAL ASK
13:41link
“If you want the link to the repo, comment scratch... just check the link in the description.”
Layers two CTAs: an early 'comment SCRATCH' bait for the free repo link, then an explicit end-card pointing to his free guide and a related video.
MENTIONED ON CAMERA
Visual structure at a glance.
