Big Idea

The argument in one line.

Two Claude skills collapse a multi-day UGC production workflow into a single reusable session: one decodes any reference video into a production-ready brief, the other drives TopView Canvas to animate consistent AI characters scene by scene.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You run paid ads for ecom brands and are currently paying real UGC creators per video.
You produce AI-generated video content and stitch together 3-5 tools by hand with no reusable system.
You are a Claude Code user who wants plug-and-play skills rather than building prompts from scratch.
You have hit the voice-drift or face-inconsistency problem when generating multi-clip AI video ads.

SKIP IF…

You need authenticity-first UGC with real creators -- this is synthetic content and may require disclosure.
You have no familiarity with Claude Code and are not willing to install skills.
Platform compliance for AI-generated ads is a hard requirement -- the video does not address disclosure rules.

TL;DR

The full version, fast.

The system has two parts. The UGC Decoder Claude skill takes any reference video plus a product description and outputs a complete production package: character image prompts, scene-by-scene Seedance 2.0 video prompts, assembly order, and a quality checklist. A second skill kicks in after character generation: paste screenshots of your generated characters back into Claude and it writes step-by-step animation instructions locked to those specific images. TopView Canvas is the execution environment -- it keeps character seeds and settings consistent across all clips, solving the voice-drift problem that plagues clip-by-clip generation.

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 – 00:37

01 · Introduction & hook

Finished electrolyte UGC ad plays as the lede. Creator promises to share the full system for free.

00:37 – 00:56

02 · Installing the Claude skills

Download zip from link, upload to Claude via Customize > Skills. Both skills installed one by one.

00:56 – 02:35

03 · Building the workflow with UGC Decoder

Drop reference video + product prompt into Claude with UGC Decoder skill active. Claude outputs full production package in ~5 minutes.

02:35 – 04:39

04 · Character generation in TopView Canvas

Use Claude-generated prompts in GPT Image 2 inside TopView Canvas. Iterate on prompts via Claude to fix polish and background inconsistency.

04:39 – 07:45

05 · Animating with Seedance 2.0

Feed character screenshots back to Claude, get per-scene animation prompts. Generate scenes 2-8 in TopView Canvas using Seedance 2.0 (CRN 2.0), 9x16, 4-5s each.

07:45 – 09:13

06 · Finishing the video

Build static image scene (GPT Image 2), generate hook scene (split-screen), download all scenes, assemble in editing software with captions.

09:13 – 11:09

07 · Pro tips & content expansion

Generate one scene first before the full batch. Save Canvas for reuse. Ask Claude for 20 more topic variations or different formats.

Atomic Insights

Lines worth screenshotting.

The UGC Decoder Claude skill reverse-engineers any reference video into a complete production brief -- scenes, scripts, character prompts, assembly order -- in about five minutes.
Voice and face consistency across multi-clip AI video is the hardest synthetic UGC problem; TopView Canvas solves it by keeping character seeds in one storyboard rather than generating clips independently.
Generating one scene before committing to the full run is the single highest-ROI quality gate -- it catches bad character images before you burn credits animating them.
Background inconsistency between AI characters is solvable by looping back to Claude with a revised prompt before animating -- catching it at the image stage costs nothing compared to fixing it in video generation.
Once a Canvas storyboard works, it becomes a reusable template across campaigns by swapping character images and product details -- the structure is the asset, not the individual clips.
After the first video, Claude can generate 20 format variations from a single follow-up prompt, turning one workflow run into a content calendar.
The second Claude skill only fires after character images are generated -- you feed it screenshots and it locks its animation prompts to those exact faces, which is why voice stays consistent.
The system works for physical products, digital products, and any brand category -- the UGC Decoder skill is product-agnostic by design.

Takeaway

The storyboard-first method fixes AI video consistency.

WHAT TO LEARN

Generating AI video clips one at a time causes voice and face drift between scenes; building everything inside a single Canvas project locks the character seed before the first frame renders.

Loading all character images into one TopView Canvas project before generating any video clips keeps voice and appearance consistent -- the seed established at image generation carries through to every animated clip.
The UGC Decoder skill value is not just the scene scripts -- it is the assembly order and quality checklist, which remove the guesswork that normally costs credits on bad generations.
Generate and review one scene before running the rest; catching a mediocre character image at scene one prevents cascading quality failures across an entire 8-scene batch.
Background inconsistency between AI characters is solvable by looping back to the language model with a prompt refinement request before animating -- catching it at the image stage costs nothing compared to catching it in video generation.
A finished Canvas storyboard is reusable across campaigns: swap character images and product details, keep the scene structure, and rebuild in a fraction of the original time.
After the first video is done, a single follow-up prompt can generate 20 topic variations in the same format -- turning one workflow run into a content calendar.

Glossary

Terms worth knowing.

UGC (User-Generated Content): Video or image content styled to look like it was made by a real consumer rather than a brand. In paid ads, UGC-style creative typically outperforms polished brand video because it feels authentic.
Seedance 2.0 (CRN 2.0): The AI video generation model used inside TopView Canvas. It animates a static character image into a short talking-head video clip given a text prompt describing movement, dialogue, and tone.
TopView Canvas: A storyboard-first AI video editor. Unlike clip-by-clip generators, Canvas stores all character images and settings in one project, ensuring consistent faces and voices across every generated scene.
Claude Skill: A packaged, uploadable prompt system for Claude Code that gives Claude a specific domain expertise. Installed via the Claude Customize menu and activated by name in any conversation.
Character seed: The source image used to anchor an AI video generation. When all clips in a project reference the same seed image, the resulting videos share the same face, voice, and visual style.

Resources

Things they pointed at.

00:35toolClaude Skills (UGC Decoder + Canvas Guide) ↗

02:14toolTopView Canvas ↗

04:45toolGPT Image 2 (via TopView Canvas)

04:47toolSeedance 2.0 / CRN 2.0 (via TopView Canvas)

Quotables

Lines you could clip.

00:46

“Nobody is showing you the actual system underneath for making AI UGC videos like this.”

Strong positioning hook, works standalone→ TikTok hook↗ Tweet quote

06:51

“I personally have not done a single regeneration throughout this entire animation build. Every clip you saw was generated on the first try.”

Credibility peak moment -- proof that the system works→ IG reel cold open↗ Tweet quote

10:35

“Once you build a workflow that works, you can literally reuse it. That's the whole point of the system.”

Tight principle statement, no setup needed→ newsletter pull-quote↗ 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.

metaphoranalogystory

00:00So I'm about to give away something that took me weeks to build. Most AI tutorials show you how to make simple videos using simple prompts, and nobody is showing you the actual system underneath

00:12for making AI UGC videos like this. Wait. So you're saying that most electrolytes

00:18we've been drinking this whole time actually do nothing? So I built two cloth skills that fix this. And I used the system for my own projects and for my clients, and results really do speak for themselves.

00:31Today, I'm sharing the whole workflow, including these two skills for absolutely free. I'll leave a link for them in the description down below.

00:38So I have my laptop over here, and the first thing you need to do is just download a zip file from the link down below. Open Clot. Go to customize, then skills.

00:47Click the plus icon and hit create skill. Click upload a skill and upload both zip files one by one. Now let's use it.

00:56So podcast style and interview style videos are absolutely crushing it right now in UGC. They work for physical products, digital products, really anything.

01:05So I found this reference video right here. I want to make something exactly like this. Two people talking in a podcast setting just for a electrolyte product.

01:14So I'll open Cloth Cowork and activate UGC decoder, which is one of the skills that you get. The reason I'm using Cowork is because it lets you to upload videos directly into Cloth, and you can just download Cloth Cowork and use it the same way as you're using Cloth on the browser.

01:31I drop in the reference video together with this prompt and click send. The only thing you need to adjust in this prompt is your product, and the skill we installed is already optimized to apply the same principles for any product you want. So it builds completely unique setup each time.

01:48Now Claude reads through the entire video frame by frame, and in about five minutes, it breaks down the whole structure. This might look overwhelming at first, but don't worry. It's actually very simple.

01:58We just follow the steps one by one. So the first thing we need is the actual characters who will be talking in our video. So I'll just ask Claude to give me prompts to generate both characters, which will then be animated into all of the scenes.

02:12Now here's where TopView Canvas comes in, and this is where the magic happens. TopView Canvas lets you create AI videos by designing the storyboard first. It feels more like working in Figma than using a traditional video editor.

02:25Everything happens in one place. Character creation, storyboard building, and the actual video generation. If you want to try this yourself, I'll leave a link for TopView Canvas in the description below.

02:34So I log in to TopView and open Canvas. Click start creating and click AI image on the left sidebar right here. Select GPT image two, set the format to vertical nine by 16, and leave the quality at two k.

02:49I dropped in the first character prompt we got from cloth and just hit generate. And just like that, we have our first character ready. Now I could animate it into a video right away by clicking generate video, but let's get the second character done first so we have everything ready before we start animating.

03:06Same process, drop in the second prompt and hit generate. Now I personally really like how the first image came out. The second one looks good, but maybe a little too polished and clean, to be honest.

03:17So I'll go back to Claude and ask it to generate a new prompt to achieve a more realistic look. I just tell that the second character came out looking too polished and ask it to improve the prompt. After we got a new prompt, just drop it into Canvas and hit generate.

03:32Alright. Much better now. But now we have a slight background inconsistency between the two characters as you can see.

03:38So I'll go back to Claude one more time and ask for a new prompt. So the first chapter appeals in a matching background. And this is exactly the kind of thing I want to show you, by the way, the real problems that come up when you build this and how to fix them.

03:51Right? I'll drop in the new prompt and this time select three generations instead of one so I have more options to choose from. And just like that, we get three new images, and now the backgrounds are matching perfectly.

04:02And never take for granted a small details like this because this is what separates mediocre AI content from ultra realistic results. I'll pick this one as our final character. Now that we have both characters ready, we can start animating them and constructing the full video.

04:17And here's why I love using Claude skills for this. Instead of guessing what to do next, I'll just drop a screenshot of both generated character images into Claude and type in this prompt. And just like that, we have a full step by step breakdown of exactly what to do to build this entire video.

04:34As you can see, step one was opening Canvas and generating the chapters, so we can move on to step two, which is generating scene two for speaker b. Here's the full prompt including the script he needs to deliver.

04:46I'll copy this prompt and head back to Canvas in top view, select the character image I want to animate, and click generate video. Drop in the whole prompt. Make sure c r n two point o is selected.

04:56Format is nine by 16. I'll set this one to five seconds and hit generate. And there we go.

05:01Our first video clip is ready. Yeah. And the reason why is something almost no one in the industry wants to talk about.

05:07This scene plays right after the hook, which we will generate in step seven. Now let's move on to the next step. Same process.

05:15Copy the prompt, select the reference character image, drop in the prompt with the same settings, and hit generate. And here's the second clip. Your cells need the right mineral ratio to actually absorb the water.

05:25Without it, the water just passes straight through you. Now for the next steps, we have scene five, six, and seven. I'm going to speed run through these scenes and generate all of them back to back using the prompts, settings, and instructions that Claude gave us.

05:39Alright. The generations are done and honestly, I'm amazed by how well these came out. Magnesium is the one people always forget and it's the one that actually helps your muscles recover and keeps you from cramping.

05:51The hand movements, the gestures, everything just feels incredibly real. And the voice actually stayed consistent, which is the beauty of using Canvas on TopView.

05:59Everything lives in one place, your characters, your scenes, your generations, all consistent and organized. Now let's generate scene eight, which features speaker a.

06:09Copy the prompt, head back to top view canvas, select the character, click generate video, drop in the prompt, the same settings, and just hit generate.

06:18And there we have the scene number eight. And it's the only electrolyte I actually drink now. Alright.

06:24We just need two more things. Scene number four is going to be static image, so we can make it with GPT image too. But I also gonna need an extra clip from speaker b, which will play on top of that image.

06:37And scene number one, which is going to be the hook. And I want to be honest there with you. Okay, guys?

06:42I personally have not done a single regeneration throughout this entire animation build. Every clip you saw was generated on the first try, and that's what the right prompting actually going to achieve for you.

06:56The amount of credits this is going to save you is insane. And imagine how much time you would have spent figuring all of this out yourself from scratch, like designing the whole video structure. And the same way we decoded this reference video and build the entire structure, you can apply this exact same process to any UGC video out there because these skills are optimized for every popular UGC format.

07:20Or if you don't have a reference video at all, what you can do, you can just open Clot once you finish installing the skills, and then you can briefly describe what type of video you want to build and construct the whole thing step by step together with Claude. And if you want to take this even further, TopView Canvas also has a built in AI agent that helps you brainstorm and manage your creative assets directly inside the tool.

07:43Now let's finish the whole thing. I will build scene four by copying this prompt into GPT image two right here on the same canvas. And while that's generating, I'll generate scene number one, which involves animating both chapters

07:57with their individual prompts. And just like that, everything is ready. Step number eight in our Claude breakdown shows us the final assembly order.

08:05So I'll just download all of the scenes, drop them into my editing software following this exact order, cut them together, and add captions. And here is the final video. Wait.

08:16So you're saying that most electrolytes we've been drinking this whole time actually do nothing? Yeah. And the reason why is something almost no one in the industry wants to talk about.

08:25Your cells need the right mineral ratio to actually absorb the water. Without it, the water just passes straight through you. The two you always hear about are sodium and potassium.

08:32And, yes, you need both. But the ratio matters more than the amount. Most drinks get this completely wrong.

08:38Magnesium is the one people always forget, and it's the one that actually helps your muscles recover and keeps you from cramping. And then we added coconut water concentrate as the base because it's the fastest natural source of electrolytes your body can actually use. That's what makes the formula work.

08:51It's the ratio and the source together. And it's the only electrolyte I actually drink now. Like, I tried everything before this.

08:57This format could literally become a series for your brand. That's what I recommend for the brands that I work with myself to to have some sort of series that you can make consistently.

09:08And once you finish building your first video together with Claude using these skills, what you can do is you can just ask Claude to give you 20 more UGC video topics for your specific product in this exact format. Or you can just get creative and say, look, I want different format. I want street interview format.

09:27I want a casual UGC format. You you can just get so creative and just keep brainstorming and generating these new ideas. This format is also very great for retargeting ads or just organic content in general.

09:39Now before you go and build this yourself, a few things to keep in mind. Always generate one scene first, just like I did myself, and check it before generating everything else.

09:50You don't want to waste credits on animating images that don't look that good, that don't look ultra realistic because then your videos is not gonna come out looking the way you want. Using the storyboard first generation using Canvas on top view can drastically improve the success rate of your AI video output.

10:09Because as you saw, when we are generating the video, the voice actually stayed consistent across multiple clips. And if you were to generate videos like that individually, there's a risk that your voice can literally be different in each clip, and then you can't really construct the whole thing.

10:27And instead of generating blindly and wasting credits, you plan everything out visually into Canvas. And once your storyboard is ready, you can generate the final video with c dense two point o directly inside the tool. Make sure you save your Canvas because once you build a workflow that works, you can literally reuse it.

10:45That's the whole point of the system because then you can just swap certain images with different prompts and continuously keep building your AI UGC content. Both Glot skills along with all of the prompts and the link for top view canvas are down in the description below. If you found this useful, consider subscribing because I'm building more of these systems and sharing them for free.

11:06Thank you so much for watching and I'll see you in the next

The Hook

The bait, then the rug-pull.

Weeks of workflow engineering -- compressed into a zip file. The opening frame shows a face-to-camera confession, the kind creators use when they are about to hand over something real: a two-skill Claude Code system that decodes any reference UGC video and then animates an AI-character version of it, start to finish, inside a single TopView Canvas session.

Frameworks

Named ideas worth stealing.

00:56model

UGC Decoder Skill

Claude Code skill that takes a reference video + product description and outputs: character image prompts, scene-by-scene Seedance 2.0 video prompts with scripts, CTA structure, pacing breakdown, assembly order, and a pre-generation quality checklist.

Steal forAny brand that produces video ads and wants to reverse-engineer what makes reference ads work, then build a scaled version with AI characters

07:40concept

Storyboard-First Consistency Method

Instead of generating AI video clips individually (which causes character drift), load all character images into a single TopView Canvas project at the start. All subsequent generations reference the same seeds, locking face, voice, and style across every scene.

Steal forAny multi-clip AI video production workflow where consistency matters

CTA Breakdown