Big Idea

The argument in one line.

A Claude design only becomes consistent once its winning traits are written down as explicit, checkable rules, and the same measure-codify-rebuild loop that fixes a single poster also holds up across a full website and a short motion graphic.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You've gotten a great one-off design out of Claude but can't get it to reproduce that exact look on the next request.
You build or sell websites, posters, or short motion graphics and want a repeatable process instead of re-prompting from scratch each time.
You're comfortable using Claude with a CLI/MCP-connected setup and are willing to try a couple of paid third-party accounts (image/video and voice generation) to follow along.

SKIP IF…

You're looking for a from-scratch design theory course; this assumes you already know what 'good' looks like and just want to make it repeatable.
You don't use Claude with developer/CLI-style tooling; several steps assume MCP connections to outside services.

TL;DR

The full version, fast.

Claude can produce a genuinely beautiful design, but asking for 'the same style' again usually returns something different, because a verbal description leaves out the specific measurements that made the original work. The fix is a two-step system: run a design through a critique loop (several reviewers each checking one constraint) until it's genuinely strong, then deconstruct that result into a written rule set covering ratios, color allocation, the one deliberate rule-break, what's deliberately missing, and named layout patterns. That rule set becomes a reusable skill, tested by rebuilding the original from the rules alone. The method is demonstrated on a poster, a Duolingo-style website with its own anti-slop copy system, and a Vox-style motion graphic with tagged, emotion-aware narration.

Free for members

Chat with this breakdown — free.

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

Create a free account →

Chapters

Where the time goes.

00:00 – 01:13

01 · The Consistency Problem Explained

Jack frames the core issue: Claude can make one beautiful design but rarely reproduces it, and introduces the 'crusade against AI slop.'

01:13 – 01:42

02 · Why Great Designs Never Repeat

A design system alone isn't enough; without a record of why a design works, Claude can't replicate it.

01:42 – 02:13

03 · The Two-Step System

States the fix: make one great design via a critique loop, then codify what makes it good into a written, reusable skill.

02:13 – 09:22

04 · Level 1: The Poster

Builds an AI conference poster from a Savee reference using the Design Loop skill and Higgsfield, then deconstructs it into a Design DNA skill via a Notion doc and rebuilds it as a reusable poster skill.

09:22 – 14:46

05 · Level 2: The Website

Recreates a Duolingo-style AI-tutoring site ('neuro') using Refero Design references and Higgsfield-generated graphics, complete with a component/character inventory and a dedicated anti-AI-slop copy system.

14:46 – 19:54

06 · Level 3: The Vox-Style Motion Graphic

Reverse-engineers a Vox explainer clip into its component assets and rebuilds a 10-second Springer Spaniel motion graphic using Higgsfield for visuals and Fish Audio for tagged, emotion-aware narration and sound effects.

19:54 – 20:10

07 · What's Next

Closes by teasing that design skill alone isn't enough without an agentic operating system, setting up a follow-up video.

Atomic Insights

Lines worth screenshotting.

A model that isn't told which details matter fills the gaps with the average of everything it has seen, and that average is what gets recognized as AI slop.
Verbally asking for 'the same style' twice routinely produces two different results, because the description omits the specific measurements that defined the original.
A structured critique loop with multiple reviewers, each checking one constraint like spacing or brand accuracy, catches errors a single overall judgment misses.
The real test of a design rule is whether it's specific enough that a bad copy would fail it; vague rules let anything through.
Rebuilding a design from written rules alone, then comparing the rebuild to the original, shows exactly which rules got left out.
Color rules should specify not just which colors are used but how much of each, since two designs can share a palette and still look completely different.
A large multi-section website uses the same deconstruct-codify-rebuild process as a single poster; the rule set just gets longer.
A reusable, named component or character inventory is what lets AI-generated output at scale look consistent instead of assembled from mismatched parts.
Copy that reads as AI-written is a bigger credibility problem on a generated website than pacing or length, and it can be fixed with its own dedicated anti-slop rule set.
Voice generation sounds noticeably more expressive when emotional intent is written into the script as explicit tags instead of left for the model to infer.
The deconstruct-and-rebuild method for design also works on video and audio: a short reference clip can be broken into component assets and rebuilt to spec.
Testing a system on an unusually hard target, instead of something you'd have gotten right anyway, is stronger proof that the method actually generalizes.

Takeaway

Consistency comes from written rules, not luck.

DESIGN SYSTEMS

A generated design only stays consistent once the traits that made it good get written down as checkable rules, and the same deconstruct-codify-rebuild loop scales from a poster to a full website to a motion graphic.

01The Consistency Problem Explained

A model that isn't told which details matter fills the gaps with the statistical average of everything it has seen, and that average is what gets recognized as AI slop.
Being able to produce one great result once is not the same skill as reproducing it on demand; the second requires a written, checkable record of what made the first result work.

02Why Great Designs Never Repeat

Asking the same model to redo a design from a vague verbal description of 'the same style' routinely produces two different-looking results, because the description leaves out the specific measurements that defined the look.
The fix isn't a better one-time prompt; it's writing the rule down the first time so it never has to be re-guessed.

03The Two-Step System

Making a design well and being able to explain in writing why it's good are two separate skills, and skipping the second one is why the first one doesn't scale.
A structured critique loop, where several independent reviewers each check a different constraint against a design, catches errors a single pass-or-fail judgment misses.

04Level 1: The Poster

A repeatable design system starts by picking apart an existing example across concrete axes: ratios, color allocation (not just which colors but how much of each), the one deliberate rule-breaking move, what was deliberately left out, and named layout patterns.
Writing rules specific enough to fail a bad copy is the actual test of whether you've captured what makes a design work; vague rules let anything through.
The rebuild-from-rules-alone step is diagnostic: whatever comes out wrong is a rule you forgot to write down, so you add it and try again until the rebuild matches.

05Level 2: The Website

The pattern that works for a static poster (deconstruct, codify, rebuild) applies unchanged to a full website; the rule set just gets longer, not different in kind.
A named, reusable component or character inventory, the same handful of assets reused with different poses or contexts, is what makes a large design output feel consistent instead of assembled from mismatched parts.
Copy that reads as AI-written is usually the biggest credibility problem in a generated site, ahead of pacing or length, so a dedicated anti-slop copy pass is worth building as its own system.

06Level 3: The Vox-Style Motion Graphic

The same deconstruct-and-rebuild method extends past static design into motion and audio: a short reference clip can be reverse-engineered into its component assets and rebuilt to spec.
Voice generation gets noticeably more expressive when emotional intent is written explicitly into the text as tags, rather than left for the model to infer from plain sentences.
The harder and more unusual the target format, the better it proves the method generalizes; testing on something outside your comfort zone is stronger proof of a system than testing on something you'd have gotten right anyway.

Glossary

Terms worth knowing.

Design Loop: A workflow where Claude generates a design and several sub-agent critics each check it against a different standard, such as spacing or brand fit, across repeated rounds until the design passes.
Design DNA: A written rule set that captures the ratios, color allocation, and layout patterns behind a specific design, turned into a reusable skill so the same look can be rebuilt on demand.
MCP: Model Context Protocol, a connector standard that lets Claude call outside tools and services, such as an image generator, directly from a chat session.
Higgsfield: A third-party AI image and video generation platform, connected to Claude here through an MCP/CLI integration.
Fish Audio: A text-to-speech platform used for narration and sound effects, with emotion tags that make the generated voice sound less flat.

Resources

Things they pointed at.

03:15toolSavee

04:41toolHiggsfield ↗

08:26linkDesign DNA skill (Notion doc) ↗

09:48toolRefero Design (styles.refero.design)

10:02productDuolingo

15:16toolFish Audio ↗

Quotables

Lines you could clip.

04:27

“If you ask Claude 10 times to improve a design, it comes back with 10 different things.”

relatable pain point that sets up the whole video's premise→ IG reel cold open↗ Tweet quote

04:54

“It guesses the average of everything it has ever seen. That average is what people call AI slop.”

tight, quotable definition of AI slop with a strong final line→ TikTok hook↗ Tweet quote

13:44

“Everyone uses AI. Almost nobody is good at it.”

punchy before/after copy rewrite example→ newsletter pull-quote↗ Tweet quote

14:32

“Is it climbing Everest? It's even harder than that, guys.”

comedic beat that sets up the hardest challenge in the video→ TikTok hook↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

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

metaphoranalogy

Claude design builds beautiful websites and motion graphics, but 99% of them are either generic or just inconsistent.

But in this video, I'm gonna show you exactly how to build beautiful designs with Claude. Building gorgeous graphics every single time.

And how to turn those into a system so you can get that incredible output each time. And this works for any design style even if you're a complete beginner.

So if haven't you already, grab that beautiful coffee and let's dive straight in. Building beautiful designs in Claude can feel like magic, but when you use the right system, like you're gonna learn by the end of this video, you're gonna feel like a magician every single time in Claude regardless of your skill level.

Now the idea is one system for every single format. Whether you want to build beautiful websites, whether you wanna go ahead and do graphics, design posters, or you just wanna go ahead and use anything like HTML emails, this system applies to everything. But before you understand what the system is, I have to explain one thing, and that is the actual problem that we are solving here.

As I always say, we have begun our crusade against AI slop and I invite you to join me with it on this channel. Now, this is called the consistency problem. There is no point building something that is beautiful and fits for purpose unless we can actually replicate it and Claude fundamentally understands what it's doing.

You can't use it for YouTube, for Instagram, for any of your clients if you can't replicate it. And a design system only goes so far.

And to do that, we're gonna use one of the strongest features inside Claw Design that most people don't even know exist to solve this issue. And the great news is we only need to do two things. Number one is make the design amazing.

And number two, actually codify what makes it brilliant such that it can reproduce this any single time that we want it to. And by the way, if you're new here, my name is Jack.

I built and sold my last tech startup with like a gazillion customers, and now I'm building my own AI startup. And on this channel, I share here the stuff that actually works. So if you want more of that, feel free to hit like, subscribe, drop a comment.

I really appreciate the support. Now step one is gonna be the design loop.

If you're not familiar with the design loop, you can check out this video right here. But essentially, what we're going to do now is use the design loop to build something. And we're gonna do this across three levels of epicness.

Start with something that's really cool and build something progressively more impressive as we go ahead and actually use it. And then once we've done that, we're gonna take our wonderful design and then we're going to deconstruct the DNA of it such that we can rebuild it as many times as we want. Thereby solving the big consistency problem that takes you from a dabbler to a full full blown design genius inside Claude.

And we're gonna do this across a poster, a website, and also we're gonna use Vox style animations so you can capture the glory and the wonder of what that looks like. Beginning with the poster. So what I'm gonna do is shoot over to this website here, which is saved.

This is like your designer's favorite kind of like location to get different graphics. It is amazing. And just looking through this, you can see how gorgeous and beautiful all the different design graphics are.

For this one, let's go ahead and design a poster. I quite like the look of this one. So let's say that we're doing a, I don't know, AI conference.

Right? And we say, well, I want this design. I wonder if I could build something similar in Claude itself, actually using this principle.

What you're do is come down and either save design or screenshot it. Again, these guys are not sponsored. I just think their website is freaking amazing.

And once you've got that, I want you to come over and grab this design loops skill. All you're gonna do is come down and download it. This will be available in the second link in the description.

I've put a copy of this inside my free community so you can literally come down and grab it. What I'm gonna do is let you control all of this, shoot over to Claude, and then just say, hey, turn this into a skill. Like so, just drop it down and Claude will turn this into a skill.

Then effectively, all you do is do forward slash and you're gonna type in design loop like so, and then then you can go ahead and use it. And if you're not familiar with how it works, effectively, what it does is you provide to Claude what the dream outcome is.

And like, I want it to look just like this and Claude will actually spin up two or three different sub agents and they all challenge Claude in a loop to make sure that it's accurate. Because if you ask Claude 10 times to improve a design, it comes back with 10 different things.

So one of the agents will say, hey, my job is to make sure that it marries everything perfectly and we have the right spacing. The second one might be ensuring it sticks on brief. And it goes around these loops until the design is perfect.

It is by far the biggest hack I found for getting incredible designs. And for generating images, we're gonna be using Higgs Field, which is a beautiful platform that lets you basically just use any image model, any video model. You can literally connect them by coming over to the MCP and the CLI.

So all you're gonna do on this page, literally guys, I'll put a link down below so you can only just click this and take it directly here. You can just come down, screenshot this if you're not connected already. And what we're going to do is basically just get caught to connect to it because we're gonna be building images and videos and we just wanna connect all this in one sequence.

So just literally screenshot this or just literally copy the link that I give you and come back over to Claude. And if you're not connected, paste it in and say, hey there, go ahead and connect to Higgs Field with the CLI. Just like that, or just say, hey, follow these instructions.

And then once you're connected, effectively, you're gonna come down and give it this prompt. I'm gonna say, hey, I want you to design for me a version of this for AI creators for an AI conference that's coming up.

And when it's done, just create me a second file explaining how all of the design critics worked in creating this. And watch that and then send it straight off.

And by the way, if this sounds like Spanish, I'm gonna put a link down below for my full Claude Code Masterclass, which is available inside the community. I take you through foundations, websites.

I'm actually building out a full Claude design system course, stuff I have never shared on YouTube ever. It is gonna be the best that I've ever seen on it. I can't wait to share it with you.

We've got design systems, compliance, maintenance, all this stuff is live, and you also get immediate access to the entire Claude code operating system, the design OS, and a bazillion other things with this.

It is literally awesome. So if you wanna get ahead, I'll put a link down below for you so you can go and grab that.

And now this is the original, and this is what we got from Claude. Let's take a look. Look at this.

Andre Kapathi, Andrew Ng. Oh my gosh. I'm not seeing my name on here.

Guys, I obviously I have got work to do. I'm a Jack Roberts YouTube channel clearly. We've got Matt Wolf on there.

We've got Lex Friedman. Very cool. I mean, look at this.

This is the original, and then this is what Claw Design. I think that's remarkable that it's been able to do that literally using this system.

And you'll notice the attention to detail. For example, let me scroll and show you this. You see this?

Can you see how that does not exceed the line height? This is something that you get a big issue with clawed. It kinda looks okay, but isn't actually because when you go into these small details, you can see look at this.

The model has correctly identified that the height must basically must mirror this thing in the bottom right and the bottom left, which is done very, very well. And it's done good attention detail here, which I think is really, really impressive. But the real measure comes is can we make this consistent?

Well, answer is yes when we give it this skill, but I've been locked away for a long time building this out so I can show what this looks like. The idea is one design, one system, any format. Again, this will be the second link down below in the description.

Now, this explains what the problem is. So you run design loop, lots of models make lots of options, one comes up beautiful, then it stops. No one writes down, why does this look good?

We don't codeify what that looks like. So effectively, what we do is we find a design that we love, we take it apart, measure it, don't describe it, argue with yourself about which bits actually matter, write down the rules as a file, write tests that a bad copy would fail, rebuild the original from your rules alone, whatever you get wrong is a rule that you forgot, and you do it again.

And effectively, this just explains everything about what makes a good design. Ratios.

For example, the golden ratio, 1.61. How much, not just what? We talk about sixty thirty ten for color theory, weird moves, refusals, looking at what's missing, shadows, icons, curves, named layouts, you name it.

We basically break down this design from like a bazillion different lenses. This, by the way, is just at the the kind of tip of the iceberg for the stuff that we actually cover inside the community, but you get the idea of what we're looking at here. As you can see, this beautiful Notion document has everything.

And then all you're literally going to do, genuinely speaking, is come up here to share, gonna come down and click on copy link, and then head over to Claude. And then in the same chat window, just drop that Notion doc in and just say, hey there, I want you to go through this Notion document, turn it into a skill, and then run this again to turn this into a reusable skill format such that I can build these movie posters out whenever I want to.

I said movie posters, I meant AI posters. It's because they're so freaking cinematic. And then go ahead and generate the same thing in two different colors.

And then when that's complete, we've got this color, then we have this one, and then we have this one. I mean, you can pick the one that you prefer, but as you can see, we're starting to get this as a consistent thing now. And then we can just save this as a skill.

So all we need to do is forward slash poster, if that's what we save it as, and then effectively, we can just create this for whatever we want to and then just drop in anything that we need. Now the really cool thing that about this skill in particular is the fact that it should come back and query you on what do you need to do it.

So for example, if it identifies to do this, I need a subject, I need a foreground, I need a background, it will ask you for those particular things. And this takes us very nicely onto level two, which is the website.

Websites are awesome. I personally sell websites for thousands of dollars. I say that not not to brag.

It's not impressive. But just so you know that I've been through the process of building these and selling them myself, just so I I I don't just yap about what looks good. I've I've done this previously.

Websites themselves, I honestly think, are one of the easiest ways to add so much value to businesses. So what we're gonna be doing here is taking a look at a website like this, which is Referral Design. It has over 2,000 AI readable design systems.

Again, they're not sponsors. I just think it looks really freaking cool, and I love what they've built here. And I'll shout out what works.

That's that's how we rock on this channel. We will have a good time, and we shout out what works. You have a gorgeous website like this, and I I freaking love Mercury, by way.

What a cool freaking product. But you can come down, you can see this, and it breaks down everything. So the idea is that we can just literally grab a URL, come back over, and share that with Claude itself.

Now one that I really like on here is a bird that may have been chasing you around a lot. It's a bird that you may be familiar with. It is, of course, Duolingo.

I love Duolingo. I've liked Duolingo for a long time, but what's very cool about Duolingo is the fact that they've got a very cool website.

So for example, let's say that I wanna recreate this. Right? And look at it.

It's not a simple website. It's pretty gorgeous. It's got animated graphics.

It's got a scroll through. It's got free fun effective. It's got different animations you can scroll through and see.

It's a cool freaking website. I mean, this is amazing. I I love the look of this website.

It's very decent. What if I wanted to build something like that? What if I wanted to build the AI for Duolingo for AI?

Well, what I would do is copy this, head over to Claude, come right here. As you can see, my Fable limit is nearing the end at the moment. Gonna come down here.

I'm gonna type in design loop like so. I'm gonna say, hey there. I want you to go ahead and build me a version of this website, but make it like basically Duolingo for AI.

I want you to go ahead and use the Higgs field CLI, essentially, so you can generate the images and the graphics. And then go ahead and use my website design system skill to go ahead and create everything that shows me the copy, the magic copy, and all that sort of stuff.

Alright. Come down like this, and then literally enter in that URL. And then you're already connected to Hixvold, so you just click send.

And this is an example I actually did in my last video, but check this out. Look at this, guys. And as you can see, we've got all these different it built all these custom graphics, all these images I built on Higgs Field the graphics as well.

I've got something I can interact with. I've got individuals flying. I mean, like, all these by the custom graphics, custom styles is amazing.

Like, it genuinely is. And look at this. I think this section's sick.

Five minutes wherever you are. Really freaking cool. NeuroPro, a portfolio, not a paper certificate in a month.

You'll prompt like you've meant it. Really freaking cool. Come down, and then we've got loads of nice things happening at the bottom, and then everything right there.

And as you can see, it is pretty similar to Duolingo. It's really understood the dynamics of it. But you might be thinking, Jack, that looks good, but how do I know it's done that?

Well, one of the design systems and I I built this inside my community specifically because I I really wanted to give everyone, like, a really cool edge. But, look, if you click on system, check this out. This is the design system behind it.

Right? And I can see all the typography. I can see basically all the buttons that we use.

And if you've ever used Duolingo and you've been harassed. There is support out there for harassment from the Duo Owl, by the way. But as you can see, you'll be very familiar with these buttons, the lesson paths, iconography, the login screens.

I mean, it's just freaking gorgeous how much it's locked this down. But that's only the beginning. The second is getting a graphical design system, which my system does, which is cool.

It basically has produced cast members. It's got rules for how to build them, and look at the asset inventory. And now I've got this asset inventory locked down, I can essentially just reproduce any ones I want to because I've got the characters nailed down.

So if somebody's reading the book or somebody's drinking some beautiful beverage or whatever they're doing, it's fine. And then, obviously, the other thing I've got here is a model sheet, which basically explains, like, the height and the size so that everything should remain in proportion along with our color palette, the motion loops.

And then crucially as well, one big thing to elevate the quality of your design is gonna be graphics. This automatically pulls down loads of fonts that you can have a look and play around with, and as well as what is considered to be recommended based on the kind of website that you're doing, which I think is really freaking handy.

And whether you give this to a client or not, do they feel like they have some optionality? It's amazing. And then crucially, we have the copy.

So you don't have AI slop. This copy system that I built here, by the way, basically draws on the best anti slop repos that exist, Wikipedia principles, a cleanse from GBT 5.6, world class copy from, like, the best copywriters on the planet, removes loads of AI garbage, built on principles like don't make me think, name the pain, and it will actually go through the entire website.

And then instead of, like, having, like, basically, like, poor converting stuff, like, get started, start through. Instead of the fun way to learn AI, it changes it to everyone uses AI, almost nobody is good at it.

And then it's just a really sick thing and just goes down that and then drastically d AI slopifies your websites.

I responded to in my community, I'm not joking here, hundreds of websites. I remember I did a big thread on this on my community, and literally the biggest feedback I had for people was not a, too much text, but the second issue that they had was just it looked like it'd been written by AI.

So this really helps with that from a sales point of view and also a general copy. But then that takes us very nicely onto level three, which is the most ambitious and difficult challenge of them all.

What is that, Jack? Is it climbing Everest? It's even harder than that, guys.

It's the Vox graphic star. Now, funny story. When I was looking at a what would really tough challenge, I saw so many saying, like, videos on Vox graphics.

I was like, you know what? We're gonna freaking do it. We're gonna have a little playing around with it and see what it looks like because I also just wanted to show you this works for literally anything.

So this is the one that we're gonna be using, this fox graphic right here. So I'm gonna hit play so you can have a little play around with it. As you can see, brings out a shark, brings it across.

Another arrow comes here. Nice narration style, four feet four inches. And as you can see, they've got a really cool narration here.

Less than a year old. That made it easier to move. Cool.

So maybe we're doing explain a bit together and I just wanna crack the Vox style. Well, how do I do that? Well, first thing I do is I'm gonna come back, and I'm just gonna grab where this exists.

So I'm gonna come down. Let's just copy video video URL at current time, and head over to Claude. And then, of course, as always, I'm gonna come down for my design loop skill.

I'm gonna say, hey there. I want you to go ahead and create for me a ten second motion graphic with in in the Vox style, and I want you to use this as your gold standard.

This this ten second clip. Okay? And I want you to go ahead and use Higgs field for the image and videos, and I want you to use fish audio for the actual audio voice on here, please.

And then go ahead, build it, and make it about a Springer Spaniel. If you don't know, I have a Springer Spaniel. He's pretty freaking cool.

I love him. He's amazing. And then we're just gonna send this one off.

Now, when I'm doing any kind of voice generation, I use Fish Audio for a couple reasons. Number one, it is about three to seven times cheaper than alternatives. But the coolest thing about this, genuinely speaking, is the fact that when we write things, we don't just wanna have it in a plain language.

And if we're doing a Vox style video, there might be times where we wanna be sad or excited. The very cool thing that you can do with this actually is you can add in tags. So I could say something like, the English Springer Spaniel is a powerful dog that loves hunting.

But unfortunately, they eat a lot of schmackers. Something like this. Right?

Then if you click on auto tag, what this will actually do is add in any, like, emotions for anything you're doing. And I mentioned this just because I want you to win. Right?

And if you're building, like, actual voice and audio in something, you want to sound good. So I might say this and again, you can add these in. You don't have to follow these ones, but you can just literally add in any emotion that you want to.

Crying loudly. I don't know. Let's play around with it.

And then we can pick our voice over here. We've the Claire Jung narrator. Let's go into more voices here.

And they have this ridiculous library, which is freaking sick. They even had one for Morgan Freeman, which I thought was funny. Claude actually refused to use Morgan Freeman.

I thought that was highly, highly controversial. But let's get down to it. We've got Mortal Kombat.

Neutrola. You get the idea. One that I really like though, I think is this one here.

Consider how often Cool. So if you're doing whatever it is, we can click on use for that, come down, and then we can come down here and generate the speech. And we're using the s 2.1 model, which I really like.

The English Springer Spaniel is a powerful dog that loves hunting. You get the idea.

You can play around with it, and we can connect this directly into Claude. So if you come down here and click the bottom left, which is developer, and then on the left hand side, you're gonna see API keys, and then you're gonna click on create API key, and then essentially, you can head over to Claude and just say, hey. I want you to connect to fish audio using the below credentials.

Let me know anything that you need. And then when prompted, just provide it with your API key, and then it can actually go ahead and build everything. And then when it's finished, it's produced this ten second clip for us.

And I'm gonna have the the vox on the background so you can see the similarities. So here's the vox.

This is the English Springer Spaniel,

20 inches tall, 50 pounds,

bred to spring birds out of cover, And that is how it got So what's interesting to me on this one as well is we even had from Higgsville, as you can see, look at this animated background. It produced a unique background and video specifically for this animation, which is cool.

And bear in mind, it's done all of this basically copying that exact style that we gave it. So if we gave it a more complicated style or a less complicated style, we could literally go ahead and replicate that, which I think is cool that you can give it this video and essentially just kind of like create anything that you want to.

There's a spaniel happening in the background. We have some leaves falling, which is awesome. 50 pounds is great.

And then we have this stuff here. But obviously, the really cool thing if we wanted to, we can also bring in if we want to sound effects from fish audio. So we're connected.

We can bring in the audio. We can add in various different things. We have the river.

We have everything we need to. And just like before, in the last design, what this will effectively do with this prompt and the system is reverse engineer what it is that makes it.

So for example, what it will do is identify all the assets like, hey, it has an arrow, it has a background with a video, it has this kind of text. And it essentially breaks down in step one what all of the different components are that make this up. So video in the background, image of the thing, and here you might say it has like a yellow explainer with a link and it goes ahead and does that.

So you can see it is whatever you essentially feed it, you can replicate it with the right systems using this prompt. Now understanding design is incredible but it's only one part of a multifaceted system.

And if you don't leverage an agentic operating system and a design operating system, you're leaving too much value on the table, which is why we're gonna learn exactly what that looks like in this video right here.

The Hook

The bait, then the rug-pull.

Most Claude-generated designs look great once and then drift the next time you ask for the same style. This breakdown walks through the fix: critique-loop your way to one genuinely good design, then write down exactly what makes it good so Claude can reproduce it, on a poster, a full website, and even a ten-second motion graphic.

Frameworks

Named ideas worth stealing.

02:13model

The Design Loop

Build
Critics
The Bar

Claude generates a design, then several sub-agent critics each check it against a specific standard in repeated rounds until it clears the bar.

Steal forany AI generation task, not just design, where a single pass/fail judgment isn't catching inconsistencies

05:25list

The 60-Second Design DNA Process

Find a design you love
Take it apart - measure it, don't describe it
Argue with yourself about which bits actually matter
Write the rules down as a file
Write tests a bad copy would fail
Rebuild the original from your rules alone
Save the whole thing as a skill

The seven-step loop for turning one good design into a reusable, rule-based skill.

Steal forany repeatable creative output where 'redo this in the same style' currently means re-prompting from scratch

05:55list

The Six Things That Make a Design a Design

Ratios
How much, not just what
The one weird move
The refusals
What is missing
Named layouts

The six axes used to pull apart a reference design into concrete, writable rules.

Steal forauditing any visual reference before trying to brief an AI tool on it

06:10concept

60/30/10 Color Rule

A named ratio for allocating dominant, secondary, and accent colors across a design, referenced as part of the 'how much, not just what' rule.

Steal forwriting a color rule specific enough that a model can't guess its way around it

CTA Breakdown

How they asked for the click.

VERBAL ASK

19:54next-video

“if you don't leverage an agentic operating system and a design operating system, you're leaving too much value on the table, which is why we're gonna learn exactly what that looks like in this video right here”

Soft cliffhanger pointing to a follow-up video on the broader 'agentic + design operating system,' rather than a hard subscribe/link ask.

MENTIONED ON CAMERA

04:41toolHiggsfield ↗

08:26linkDesign DNA skill (Notion doc) ↗

15:16toolFish Audio ↗

FROM THE DESCRIPTION

PRIMARY CTAWhere the creator wants you to go next.