Big Idea

The argument in one line.

Real ad agencies lose to one person with Claude Code not because of image quality, but because the AI faithfully mirrors the strategic brief-building process that professional art directors run before touching a single pixel.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You want to produce client-quality ad deliverables without hiring a creative team.
You already use Claude Code for development work and want to see it applied to visual production pipelines.
You run a freelance design or video business and need a repeatable, one-command campaign workflow.
You are exploring how third-party MCPs like Higgsfield extend Claude Code into image and video generation.

SKIP IF…

You need final output approved by a senior creative director before any client sees it.
You have no product image to anchor the workflow; the entire pipeline assumes a single source photo.

TL;DR

The full version, fast.

The central argument is that weak AI ads come from skipping strategy and going straight to image generation. Ad Studio, a Claude Code custom skill, enforces a seven-phase pipeline: brand research, concept selection, hero casting, visual DNA definition, storyboard with a four-act story arc, frame generation via the Higgsfield MCP, and final video render via CeDance. Each phase passes its output into the next, so the final 15-second clip is anchored to a real creative brief. The presenter runs the full pipeline live on a North Face yellow jacket, from a Google Images product photo to a rendered video ad with the tagline Made for Mountains. Worn Everywhere.

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 – 00:31

01 · Hook: AI ad reveal

Red Bull surfing ad, North Face street shot, and Nike Jordan comic brief shown as proof of concept.

00:32 – 01:07

02 · The One-Person Ad Studio

Agencies needed 8 people; Claude Code collapses that to one. Presenter credentials (Apple, PlayStation, Nissan) established.

01:08 – 01:35

03 · Real campaign examples

Three full briefs shown: Red Bull POV GoPro, Nike Jordan comic style, Claude Code Club 3D motion graphics.

01:36 – 02:04

04 · Why research comes first

Most people skip strategy and go straight to image generation. A real art director spends time on buyer, style, hero, look, and story before anything is rendered.

02:05 – 03:07

05 · Full pipeline overview

All 7 phases explained: research, concept/style, hero casting, visual DNA, story, image prompts, motion and music.

03:08 – 03:25

06 · North Face demo begins

Presenter picks the North Face Summit Gold jacket; searches product image on Google.

03:26 – 04:10

07 · Setting up Claude Code

Creates project folder, opens Claude Code desktop app, selects Opus 4.8, invokes Ad Studio skill, saves product image to folder.

04:11 – 04:49

08 · Running the pipeline

Selects 16:9 aspect ratio; pipeline fires three research sub-agents in parallel.

04:50 – 05:29

09 · Brand and audience research output

Live brief HTML shows true value, ICP (18-32 fashion/street), design DNA (Summit Gold/charcoal), and off-brand blacklist.

05:30 – 06:22

10 · Reviewing research brief and concept options

Three concept options generated. AI recommends option C: Same Jacket, Different Mountain.

06:23 – 07:05

11 · Selecting ad concept C

Chooses the city-vs-mountain contrast concept. Every dramatic mountain shot cuts instantly to the boring city equivalent.

07:06 – 07:31

12 · Mood board and hero casting images

Pipeline generates mood board and AI hero images in three takes. All assets displayed in brief HTML.

07:32 – 08:45

13 · Casting yourself as hero

Presenter uploads reference photo; pipeline generates consistency token and 3 hero takes. Locks take 3 (clean front/concrete wall).

08:46 – 11:23

14 · Visual DNA and storyboard

Five-dimension look spec defined. 4-act arc written as prose, broken into 12-shot list, Claude critiques its own list. Frames rendered. Two shot changes requested before approval.

11:24 – 13:19

15 · Rendering and reviewing the final ad

CeDance called with timestamped master prompt. Returns one 15-second video with all cuts. Final tagline: Made for Mountains. Worn Everywhere. Outro and CTA.

Atomic Insights

Lines worth screenshotting.

The reason most AI ads look random is not the image model, it is the absence of a strategic brief before generation starts.
Claude Code can run three research sub-agents in parallel on a single brand, returning true value, ICP, design DNA, and off-brand blacklist simultaneously.
Defining a character in a 2-word archetype plus 3 repeatable descriptor tokens and appending those tokens verbatim to every prompt keeps a generated person visually consistent across 12 separate frames.
You only need to define five visual dimensions once and embedding that vocabulary in every image prompt makes unconnected renders feel like a single coherent shoot.
A four-act narrative arc written as prose first produces a stronger shot list than describing individual images directly to a generation model.
Having the AI critique its own shot list like a film editor before any frames are rendered catches weak or redundant shots that would waste generation credits.
CeDance accepts timestamped scene descriptions in a single prompt and returns one 15-second video file with all cuts already baked in, no manual assembly required.
The one-person ad studio is not a shortcut to quality; it is a shift in where the creative thinking happens, from the production stage to the brief stage.
Agencies historically spent the brief-building budget on people; Claude Code moves that spend to API tokens and returns the brief faster.
A 9-dollar-per-month community that ships the skill and the prompts is a distribution model as much as it is a product, the tutorial is the funnel.

Takeaway

Brief quality is what separates AI ads that convert from AI ads that just look generated.

WHAT TO LEARN

The failure mode of AI-generated advertising is not the image model, it is the absence of a real creative brief before the first prompt is sent.

Brand research should answer three questions before any image is generated: what does the brand actually sell beyond the literal product, who really buys it, and what creative directions would immediately read as off-brand.
Framing a product in a creative angle, a written concept with a point of view, produces more coherent imagery than describing the image you want directly to a generation model.
Character consistency across AI-generated frames comes from a fixed set of descriptor tokens repeated verbatim in every prompt, not from hoping the model retains visual memory between separate generations.
Defining five visual dimensions once (lighting, palette, film stock, lenses, depth of field) and appending that exact vocabulary to every generation prompt is what makes separately rendered frames feel like a single coordinated shoot.
Writing a four-act story arc in prose before building a shot list forces narrative logic into the storyboard; having the model critique its own shot list as a film editor catches weak frames before any credits are spent rendering them.
Some video generation models accept timestamped scene descriptions within a single prompt and return one assembled clip, understanding this capability changes how a storyboard should be structured upstream.

Glossary

Terms worth knowing.

Ad Studio: A Claude Code custom skill (slash command) that runs the full seven-phase ad production pipeline from a single product image, outputting an HTML brief plus rendered video assets.
Higgsfield: An image and video generation platform (higgsfield.ai) that offers a Claude Code MCP, allowing Claude to call image and video generation models directly inside a coding session.
CeDance (Sea Dance): A video generation model that accepts a timestamped scene description in a single prompt and returns a complete multi-scene video clip with cuts already assembled.
Consistency token: A fixed set of character descriptors, archetype, wardrobe, and 3 keywords, repeated verbatim in every image generation prompt to keep a subject visually identical across all frames.
Visual DNA: A five-dimension style spec (lighting, color palette, film stock/texture, camera lenses/angles, depth of field) defined once and appended to every image prompt to ensure visual coherence across a campaign.
Design DNA: Brand-level color, aesthetic, and tone attributes extracted from the product image and brand research, used to constrain the visual DNA for a specific campaign.
ICP (Ideal Customer Profile): The specific demographic and psychographic description of who actually buys a product, used in ad research to ensure the hero, tone, and creative angle speak to the real buyer.
Off-brand blacklist: A list of creative directions, visual styles, or messaging angles that brand research identifies as likely to alienate the core buyer or misrepresent the product.
4-act story arc: A narrative structure for short-form video ads: setup establishes character and context, build creates tension, shift delivers the turn or joke, resolution lands the brand moment.

Resources

Things they pointed at.

04:50toolHiggsfield ↗

11:24toolCeDance (Sea Dance)

12:53productClaude Code Club ↗

03:26toolOpus 4.8

Quotables

Lines you could clip.

00:17

“Agencies used to put eight people on an ad campaign like this. Now Claude Code has unlocked the one-person ad studio.”

Punchy thesis, specific number, no setup needed→ TikTok hook↗ Tweet quote

01:50

“Strategy first. Pixels last. That is the whole secret.”

Three-word hook shown on screen, zero context required→ IG reel cold open↗ Tweet quote

09:25

“You define the look once, in five dimensions, and each one ends with a keywords line: the exact vocabulary passed into every image prompt.”

Practical instruction with a clear mechanism→ newsletter pull-quote↗ Tweet quote

11:38

“One prompt with a timestamp per scene returns the whole 15-second ad as a single clip, cuts included.”

Counterintuitive capability, stops the scroll→ 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.

00:00These ads were made entirely with Claude code from a single prompt. Up until recently, this wasn't even possible.

00:12But now it is.

00:17Claude code is a cheap code for creating AI ads. Agencies used to put eight people on an ad campaign like this, but now Claude code has unlocked the one person ad studio. You can now create an entire ad campaign using QuadCode from a single product image.

00:32In this video, I'm going to walk you through the custom skill that I built that not only creates the ads that you saw in the beginning, but creates this entire research brief. This is great whether you're working on a project for yourself or a client that you're trying to impress. If you're wanting to learn how to use Cloud Code to make cinematic ads like this, I'm also going to walk you through the entire strategy behind it including all of the prompts that I use to make high quality ads every single time.

00:55If we haven't met yet, my name is Duncan Rogoff. I'm a former art director at brands like Apple, PlayStation, and Nissan, and I now run one of the top communities for learning Cloud Code and building income. So focus in, close all your open tabs, and let's build.

01:08So this is the ad campaign that I built for Red Bull, and this is exactly what the custom skill produces. This is a brief that I made for the Nike Jordan ones in the style of a comic book. This is the brief for a full three d motion graphics ad I made for the Quad Code Club.

01:22If you wanna get access to the custom skill already built for you along with any of the prompts we go over today, just check the link in the description, join the community, it's only $9 to get started. So what's pretty cool about these systems is they're extremely versatile and allow you to create all different styles of ads.

01:36So if you really want to create a campaign that is actually effective, most people skip this first part, and that's really the research and the strategy. And so a lot of people will just go straight to the image generator, and then they'll wonder why their ads aren't working. But a real art director is going to spend time on like who the buyer is, the style that's going to resonate with their audience, who the hero of the ad is, what the look is, and what the overall story is.

01:58So you really need to understand these components if you want to have a compelling ad that actually converts people. So today we're going to cover all of the sections that go into these ads. Research on the brand, the angle of the ad, who the hero is, what the look is, the story, any of the image and video prompts, all of the motion and music.

02:13And then at the end, I'm actually going to show you one master prompt for the skill that creates all of this from a single product image. So this is what we are going to be creating today. You already saw the ad in the beginning for Red Bull, but the skill also creates this entire ad campaign along with you.

02:28The first thing that it's going to do is it's going to research the product and the brand. It's going to research the audience. So who was actually buying this product?

02:35So how can we speak to them directly? It's going to create all of the design DNA. So what is the look and feel of this product?

02:42And then is there anything that we should absolutely avoid? From there, it's going to come up with three different concepts for you to choose from. It's going to give you options for casting the hero as well as a mood board with beautiful images like this.

02:54This isn't real, this is all AI generated. Same with this one here. It's going to decide the visual DNA to make sure that we have consistency across the whole ad.

03:02It's gonna create the entire storyboard for us. And then of course, it's going to stitch everything together in one final polished piece. So let's kick this thing off and get started while we talk about the strategy.

03:11I thought it would be fun today to try doing this for North Face. I love this bright yellow jacket. First things first, I'm just going to open up Quad Code.

03:17I'm inside of the desktop app and I'm just over here in code and you can click new session if you need a new session. The next thing you need to do is to make sure that you're inside of a project folder. So I'll just come to the desktop, I'll create new folder, and I'll just call this North Face to make sure that we are working and all ready to go.

03:33And the last piece of this is because this is relatively complex and will require multiple steps, you're just gonna wanna make sure that you're on Opus 4.8 which is really good for handling these longer sessions. So So I'm just going to type in ad studio here to bring up the scale, and then all I have to do is save this image inside of our folder.

03:48So I'll go ahead and click save, come to the desktop, go into our folder here, and I'll just save this as North Face jacket. And I'm just gonna come back to the site here, and you don't really even need to do this, but I think it's helpful to at least copy the name. And if there is any information about the product, like this little description here, which we may not even need, but it's worth it to include it so we have everything.

04:07So I'm gonna send this off and the system should come back to us with a couple of questions. So we can see that this is kicking off the Ad Studio pipeline. It's gonna lock up the format and it's going to start the research for us.

04:18So I'm just gonna choose 16 by nine widescreen because this is YouTube, but you could make this nine by 16 or vertical for Instagram Reels, TikTok, or YouTube Shorts. And basically, the first thing the system is doing is it's performing this research. So we're just giving it the product and telling it to research the brand.

04:32We want to figure out who the brand is in one line, what do they really sell just beyond the product, like what it's going to resonate with their core buyer. Who is the target audience? What is the aesthetics?

04:43So what does this actually look like? The mission of the brand, the core messaging, and then is there anything that is totally off limits that we should just completely avoid? So we can see that it went ahead and figured out what the color palette already is for us.

04:55So it's the summer gold and charcoal, which we already have. And then it's actually launching three research sub agents so that it can do all of this research at the same time. And if you're wondering why this says Higgs field, I'm going to be using Higgs field to generate all of our images and videos.

05:07They're not a sponsor of this video. I just think they have really great tools. Higgs Field has access to all the latest image and video generation models.

05:14What's also really cool is that they have an MCP that's really easy for you to connect with Cloud Code. But if you wanted to do this through a different image or video service, you absolutely could. Okay.

05:24So this is where it gets fun. Claude Co built all of this for us and it's just going to get better from here. It's pretty cool that it builds this directly inside of the app.

05:33It does sometimes have a hard time bringing the images and videos, but this is an HTML. So you can just open this up like in Google Chrome or something like this to get access to all of the images and videos. And if we have any little overlapping things like this, just tell Cloud to fix it.

05:47It's really no big deal. But here is the important part, which is the research that we just did. So why do people buy this?

05:53People don't buy this for the rain. They just buy it because it's a famous jacket, and basically, it looks good. Wearing it says, I have good taste.

06:00It's funny because it looks like serious mountain gear, which we know it is, but actually most owners just wear it around the city. So who buys it? Mostly young people, roughly 18 to 32 into fashion and street style.

06:11How should everything look? We know the color is yellow, black, and white against the gray city concrete, and then any of the ad ideas that we should avoid. So then it went ahead and gave us a couple of different options for the overall ad concept and the style.

06:24And it even goes ahead and recommends one of the options for us. So option a is the concrete summit. It's kinda funny.

06:29Your mountain top is really just top of the stairs on your street, but the one that they selected is same jacket, different mountain. So every dramatic mountain shot instantly cuts to the boring city version of the exact same moment. Same pose, same yellow jacket, and the person knows it's funny.

06:44So it takes the most overused outdoor ad shots of the hero on a mountaintop, and it just makes a joke out of it. So climbing a snowy ridge cuts to climbing the subway stairs. So it's just kind of the real life truth about who wears this jacket.

06:57It's funny, it's honest, and it's very easy to share. Let's rock with this today. So we can just come back in the quad and ask us which option do we want, and then all we have to do is say option c.

07:06So this finished up and it gave us a mood board and a couple of the hero takes for us to choose from. So it just created this beautiful snowy mountain range contrasted with the urban environment, rendered out a close-up of this jacket, like the details on this are insane. And then a couple of different options for the hero image.

07:24Now what I thought would be interesting is at this point is like, can I actually put myself into this? And I imagine that I can. So I'm going to update this with a photo of myself.

07:33So I'm just gonna go ahead, click add photos and files. So if you have a photo of yourself, like I just have this one here that you've probably seen on my YouTube a bunch, I'm gonna add this in. And can we just say, can we make this man the hero?

07:45And I'll come back when that's updated. So what the system is doing is actually defining what it looks like to put me into the ad. So all I've done is given it a clear photo of myself and all that matters now is consistency.

07:58So here's the type of prompt that you could use if you wanted to do this yourself. So here's my reference photo and give it the photo. I am the hero of this ad.

08:06So based on the brief, define my character. So the exact words that will appear every single time in every prompt to make sure that I stay the same, and then it's going to write an image prompt that casts me in this world rendered in the style that we picked in step two. So the whole idea of the system is to build off of the pieces that come before it.

08:25So this is pretty funny. It went ahead and recast me as the hero of this ad. Here's me in the city and here's me on top of the mountain.

08:33The next thing it does is it just asks me which take do I wanna lock in. Take three was definitely the best, the one of me in front of the concrete wall. So I'm just gonna go ahead and say take three, and once this is done, it's actually going to create a full storyboard for us.

08:46So there are two pieces that go into the storyboard. One is the look, and two is the actual story. So what is the story we are trying to tell?

08:54So what is the look? We don't want this to be random. We basically want to define the aesthetic, things like the lighting style, the color palette, like what is this shot on?

09:01Is this shot on film or DSLR or an iPhone? What are the camera lenses and angles? What does the depth of field look like?

09:07So any of these photography terms that can help define the look of this piece. What's cool about Quad is that it's pretty smart and based off of the product image, it's going to figure this stuff out for you. Arguably, one of the more important pieces because we're just doing this with a fifteen second ad is the story arc.

09:23What's the four act structure? So we create tension in the beginning, we have some sort of buildup or turn, we have the payoff, and then we have the brand reveal at the end. So you could use a prompt like this.

09:33Write this ad as a story. A four act narrative arc in prose, setup, build, shift, resolution, establish the character, the location, and the emotional thesis. Break it into a shot list for a fifteen second spot, eight to 12 shots.

09:46This part is cool because then we're just gonna have it critique its own shot list like a film editor. So this went ahead and finished our storyboard. It locked in the visual DNA, what this is going to look like on every shot, and it went ahead and even rendered all of our frames for us.

10:00It's pretty cool because it put me in all of these shots. It looks pretty good. You can see there's a lot of consistency in the way that it looks and the color palette, which is pretty awesome.

10:08And this is really the last chance for you to sign off on the creative direction or any of these scenes before it actually goes to create the ad. And so this is your chance to act like a creative director. Is there anything that you wanna change that doesn't make sense?

10:21For me, like this shot, like there's a person in the back, we'd wanna get rid of that. So here, maybe there's more of an opportunity for it to show me like on top of the mountain and then like on top of the staircase or something like that. But I kinda like the side by side of the hand gripping the mountain here versus the hand on the subway.

10:36Like, I think that's kinda funny. So at this point, you would just tell Claude what it is that you wanna change. And so I would say something like shot number two has an extra pair of legs in it.

10:46Shot four probably needs to be me on top of the mountaintop, juxtaposed with me on top of the staircase. Kind of the same with shot 10 of 11. I think we could show something that more appropriately shows the comparison between mountain and city.

10:58One thing I really actually did like was this end shot of the brand, this nice close-up on the jacket and the logo, and this made for mountains worn everywhere tagline. Like, I think that's pretty awesome. So this finished up and made a couple of changes that I asked for.

11:10Like, it removed that second pair of legs back here, and it went ahead and added these two new shots. One, the back of the jacket, me on top of the mountain, paired to me on top of this building in the city. So once you are happy with this, the only thing left to do is tell it to render the video.

11:24So I'm just gonna go ahead and say approved. So what this is going to do is it's actually going to call Cdance, which is one of the latest video models to render the whole shot. And so just so you understand what's going on, what's pretty cool about using Cdance is you can render a fifteen second long clip, but you can actually feed it time stamps to say, like, between zero and three seconds, do this.

11:44And then between three seconds and ten seconds, do this. So you can actually have Sea Dance include all of the different scenes and cuts for you in a single clip. Basically, was just going to take all of the storyboard images that you created and write a single long master prompt that it's going to give to Sea Dance.

12:01So at the end, you're going to get one fifteen second long video file with all of the different scenes cut together. At this stage, if you wanted to give it more direction on how the camera should move, do you want zooms, do you want pans left and right? Do you want crazy angles?

12:16Like, it's really up to you. You can always give Claude more information if you have some sort of vision in your head that you wanna achieve. So our ad just finished.

12:25Let's check it out.

12:41Honestly, that's pretty awesome. I gave it very little direction. I think you can see from here how with a little bit of additional prompting, you could get something nearly perfect.

12:49I'd probably change the music and maybe one or two of the shots, but generally speaking, I am pretty happy with this considering how little effort I actually put into building the thing. What's also cool is that it saves all of the files and the images and the videos and the research that it did in the folder on your desktop if you wanna access them for any future projects.

13:06If you wanna get access to the skill and any of the prompts that we cover today, just check the link in the description to join the community for only $9. If you wanna see how I use Cloud Code to build my own agentic OS, just check out this video right here. I'll see you over there.

The Hook

The bait, then the rug-pull.

The reel opens on a first-person GoPro shot inside a breaking wave, a Red Bull can held out against the spray, the caption reading THESE ADS. Before the viewer has a second to process the production quality, the presenter's voice lands the claim: made entirely with Claude Code, from a single prompt.

Frameworks

Named ideas worth stealing.

01:50concept

Strategy first. Pixels last.

The core thesis: most AI ads fail because people skip research and go straight to image generation.

Steal forAny AI creative workflow, establish the brief before prompting any generative model

02:05list

The 7-Phase Ad Studio Pipeline

Brand research
Concept + style
Hero casting
The look (5 dimensions)
The story (4-act arc)
Image prompts
Motion and music

A sequential pipeline where each phase passes context into the next, culminating in a complete 15-second video ad from a single product image.

Steal forBuilding any AI-powered creative production workflow or client deliverable system

08:46list

Five Dimensions of Visual DNA

Lighting style
Color palette
Film stock and texture
Camera lenses and angles
Depth of field

Define these five once, append the exact vocabulary to every image prompt. Repetition of the same keywords makes separate renders feel like a single shoot.

Steal forAny multi-image AI generation project requiring visual consistency

07:32model

Consistency Token Pattern

One clear reference photo plus 2-word archetype plus wardrobe plus 3 repeatable keywords. These tokens appear verbatim in every generation prompt to keep a subject visually identical.

Steal forAI ad campaigns, social content series, or any project with a recurring human subject

10:10model