Big Idea

The argument in one line.

Claude Skills plus GPT-image-2 and a live trend-data connector can replace most of a short-form creator's production pipeline — thumbnails, content research, carousels, animations, and scripts — cutting tasks that used to take hours or cost hundreds of dollars down to minutes.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You run a YouTube, TikTok, or Instagram channel and personally handle (or outsource) thumbnails, carousels, or short-form scripts.
You already use Claude and are comfortable installing custom Skills and pasting prompts between tools.
You want a repeatable prompt system for turning a content idea into a finished asset (thumbnail, carousel, animation, or script) instead of starting from a blank page each time.
You're open to paying for a third-party trend-data API (Virlo) to get real view-count-backed content ideas rather than generic AI suggestions.

SKIP IF…

You don't use Claude or have no interest in Skills/custom connectors — the entire workflow depends on that ecosystem.
You need fully free tooling — GPT-image-2 and Virlo both cost money on top of a Claude subscription.
You're looking for long-form video editing or scriptwriting for anything other than short-form/social content.

TL;DR

The full version, fast.

The creator built five Claude Skills — thumbnail designer, a live-data connector to a trend-tracking tool called Virlo, a carousel outline generator, a motion-graphics animator, and a script writer — and gives them away as a free download. The core mechanism: Claude drafts a detailed prompt (for a thumbnail, animation, or carousel slide) using reference images and a video idea, that prompt gets pasted into GPT-image-2 (or Claude itself for animations) to generate the actual asset, and the creator iterates by going back and forth with Claude to refine the prompt until the output looks right. Virlo supplies the raw material — real view-count data across 4M+ short-form posts — so every content idea, hook, and carousel is grounded in something that already proved it performs, instead of generic AI brainstorming. The practical takeaway is a repeatable idea-to-asset pipeline: find a proven idea via trend data, then fan it out into a thumbnail, a carousel, an animation, and a script using the same Skill-plus-prompt pattern each time.

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 – 00:17

01 · Introduction

Creator shows off finished thumbnails, motion graphics, and a carousel, says he built Claude Skills to produce all of it, and offers the skill files free in exchange for a like.

00:17 – 00:36

02 · Installing the Skills

Walkthrough of downloading the skill ZIP, extracting it, and uploading each file into Claude's Skills settings one at a time.

00:36 – 02:11

03 · Stunning YouTube Thumbnails

Activates the Thumbnail Designer skill, feeds it reference thumbnails plus a video title, gets a complete GPT-image-2 prompt back, generates a first draft, then iterates with Claude on the prompt until landing on a final style — with and without the creator's face.

02:11 – 03:13

04 · Research Content with Claude

Contrasts a low-performing post made without research against a 600K-like post made with research, then connects Claude to Virlo (a live trend-data API covering 4M+ short-form videos) via a custom connector so Claude's content ideas are grounded in real performance data instead of generic guesses.

03:13 – 05:37

05 · Carousels That Stop the Scroll

Takes a Virlo-sourced content idea, uses a carousel skill to generate a full slide-by-slide outline, then uses a GPT-image-2 prompt skill with 10-20 reference photos (plus face references where needed) to generate photorealistic AI carousel images, assembled into a finished Instagram carousel.

05:37 – 06:55

06 · Animations From Your Script

Activates the motion-graphics animator skill, drops in a line from the video's own script, gets back a ready-to-use prompt, and generates a cinematic HTML animation with Opus 4.8 in a few minutes — in both horizontal and vertical formats.

06:55 – 08:14

07 · Craft Engaging Scripts

Feeds the Virlo-sourced content idea into a script-writing skill that first generates ten scroll-stopping hook options, then expands the chosen hook into a complete short-form script with hook, story, and call-to-action.

Atomic Insights

Lines worth screenshotting.

Feeding Claude real reference thumbnails from your niche (or an unrelated one) works better than describing a style in words — it gives the model a concrete visual direction.
Generic AI content advice fails because the model has no data about what's actually performing; connecting Claude to a live trend-data source turns vague suggestions into specific, provable ideas.
A single social post that underperformed versus one that hit 600,000+ likes shows research, not luck, is usually the deciding factor between the two.
Splitting the workflow into 'Claude writes the prompt, GPT-image-2 renders the image' produces more controllable results than asking one model to do both reasoning and image generation.
Carousel images that look like real photography can be entirely AI-generated once you feed the image model 10-20 reference photos matching the desired vibe.
A cinematic motion-graphics animation built from a single script line can be produced as a self-contained HTML file in under three minutes, then converted to video by screen recording or an HTML-to-MP4 converter.
Turning a proven content idea into a full script becomes fast once you generate ten scroll-stopping hook variations first and only then commit to writing the rest of the script around the strongest one.
The same underlying idea (one proven hook backed by real view-count data) can be fanned out into a thumbnail, a carousel, an animation, and a full script — one piece of research doing the work of four separate assets.

Takeaway

One proven idea can generate four different assets.

WHAT TO LEARN

Grounding content decisions in real performance data, then reusing one Claude-plus-image-model prompt loop, turns a single validated idea into a thumbnail, a carousel, an animation, and a script.

03Stunning YouTube Thumbnails

Feeding an AI model concrete reference images works better than describing a visual style in words — it gives the model a fixed target instead of an ambiguous one.
Generating multiple versions and iterating on the prompt itself (not just the output) is what produces a final result worth using.

04Research Content with Claude

Generic AI advice about what content to make is a symptom of the model having no performance data, not a limit of the model's reasoning — the fix is connecting it to a real data source.
A post's outcome (600,000+ likes versus near-zero) can hinge more on whether the idea was researched than on execution quality.

05Carousels That Stop the Scroll

Photorealistic AI images improve significantly once you supply 10-20 real reference photos that match the intended vibe, rather than relying on a text description alone.
Splitting a workflow into 'reasoning model writes the instructions, specialized model executes them' produces more reliable output than asking one model to do both jobs at once.

06Animations From Your Script

A single line of existing script text is enough raw material to generate a finished animated asset, without writing new creative copy.
Building an asset as a self-contained file (rather than inside a proprietary editor) keeps the option open to convert it into video however is most convenient.

07Craft Engaging Scripts

Generating multiple hook variations before committing to a full script surfaces a stronger opening than writing the first draft top-to-bottom.
The highest-leverage step in a content pipeline is the research step — one validated idea can be repurposed across multiple formats (visual, written, animated) instead of researching each format separately.

Glossary

Terms worth knowing.

Claude Skill: A custom, reusable instruction set uploaded into Claude that gets activated by name and repeatedly performs a specific task, like drafting an image-generation prompt in a fixed format.
GPT-image-2: OpenAI's image-generation model, used here as the actual renderer for thumbnails, carousel photos, and reference-face composites after Claude writes the prompt.
Custom connector: A Claude feature that links the model directly to an external live data source (here, a trend-tracking API) so Claude's answers pull from real-time data instead of its training knowledge.
Virlo: A third-party paid tool that tracks over 4 million short-form videos across TikTok, Instagram, and YouTube Shorts and surfaces which ones are performing well and why, via an API key connected to Claude.

Resources

Things they pointed at.

00:00toolClaude Skills toolkit (free download) ↗

04:29toolVirlo ↗

01:31toolGPT-image-2

06:20toolClaude Opus 4.8

Quotables

Lines you could clip.

04:29

“If you ask Claude what's performing best in your niche, you just get generic advice.”

sharp, honest limitation callout before pivoting to the fix→ TikTok hook↗ Tweet quote

06:55

“A year ago, this would have cost me hundreds of dollars to outsource. Now I just activate the skill and it's done.”

concrete cost-of-outsourcing comparison, strong pull-quote→ 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.

story

00:00What if I told you that cloth just help me make these thumbnails in couple of minutes? Same with these motion graphics and this Instagram carousel. I spent weeks building cloth skills that make all of this possible.

00:11Today, I'm giving away all of them for free, and all I ask in return is for you to drop a like on this video. First, let's get these skills installed. So just grab the zip file from the link below and extract it.

00:22Then go to cloth, go to customize skills, hit the plus icon, and select create skill. From there, click upload and add each file one at a time.

00:32Takes about thirty seconds and that's it. Now let's put them to work. A good thumbnail is the difference between a video that gets views and the one that gets ignored.

00:42So I built a cloth skill that helps you generate thumbnails that look like this. So first, let's activate the thumbnail designer skill by typing this. Let's say I run a geography channel and my next video is about the top 10 countries to visit in Europe.

00:57All I need to do is add thumbnails that inspire me visually. You can find thumbnails from the same niche or from a completely different one. Doesn't matter.

01:05You are just giving cloth a visual direction to work from. Then add this prompt and drop your video title or idea into these brackets right here, and we get a complete prompt ready to go. We take this prompt into GPT image two, which is free to use in your browser, and drop it in.

01:22And a few minutes later, we get our first version. Now you don't have to stop at the first result. Personally, I always go back and forth with Claude asking to make adjustments to the prompt.

01:31So I spent a few minutes tweaking the thumbnail prompt with Claude and generating dozens of versions across completely different styles until I landed on the one I liked. And I think all of these came out really great. Now if you want your face in the thumbnail, go back to cloth, drop in the inspiration thumbnails, and use this prompt instead.

01:48Again, add your video title or just an idea into these brackets. Now when you drop this prompt into GPT image too, make sure to also attach reference images of your face. And we get this.

01:59And just like before, you can just keep tweaking the prompt together with Claude until you land on the exact version that you want. So using this one skill, you can apply this to any niche, generate and refine thumbnails in minutes. So the hardest part of being a content creator is knowing what content to make.

02:15For example, I posted this without doing any research, and as you can see, the results were not so great. Then I spent some time doing research, made this post instead, and it got over 600,000 likes. And I've got plenty more performing just as well.

02:29But if you ask Cloth what's performing best in your niche, you just get generic advice. Now we can fix this by connecting Cloth to Verlo. Unlike the skills we just installed, this one connects directly to Cloth as a live data source.

02:41And Verlo tracks over 4,000,000 pieces of short form content across TikTok, Instagram, and YouTube shorts in real time. Once it is connected, you just ask and Cloth gives you the most accurate viral content ideas. It takes less than a minute to set up, and here's how to do it.

02:56Go to the first link in the description and log in to Verlo. Head to API keys, generate a new key, then open Clot. Go to settings, connectors, add custom connector, paste the Verlo URL, hit connect, paste your key, and authorize.

03:10That is it. Now watch what happens. I just type this.

03:13Using Verlo, find me trending content ideas in the productivity niche. For each one, give me the hook, the main angle, the view count, and exactly why it performed. And in couple of minutes, we have some of the top performing content ideas in the productivity niche.

03:27These are real posts with real view counts and real breakdowns of exactly why each one worked. Take this one, for example, a split screen video contrasting bad habits with disciplined ones. No spoken hook, just visual storytelling.

03:391,270,000 views. And right here, Verlo tells you exactly why it worked.

03:44Down below, you'll also find the biggest patterns across all the top performers and what's actually working best in this niche right now. With Verlo, you can track every single major niche. I can just switch from productivity to fitness and get a stack of best performing content ideas in that niche instead.

04:00You can top up your Verlo balance for as low as $10 and run your first content research session in the next minute. I personally use this all the time whenever I'm doing my own content research. The link for Verlo gonna be in the description down below.

04:13So now we have a proven content idea with real data behind it from Verlo. Let's turn this into a carousel post like this. And none of these pictures are real.

04:21They are all generated using AI. Let me show you how. First, we turn Verlo idea into a full slide outline.

04:28Let's activate a carousel skill and drop this prompt. Just grab the content idea that Verlo gave us and put it into these brackets right here. And just like that, cloth gives us the full outline slide by slide.

04:40I tweaked it a little bit and went back and forth with cloth until the structure felt right. Now the question you're probably asking is how do I make these photos look so real? This is where the GPT image two prompt skill comes in.

04:52Find 10 to 20 reference photos that match the vibe you're going for, activate the skill, and drop all your reference images into cloth with this prompt. And if you don't have reference images, just ask cloth to generate image prompts that are related to each slide that it generated for you. Then just take those prompts straight into g p t image two and drop them in one by one.

05:14For any slide that includes a person, just attach your face reference images alongside with that prompt. After about thirty minutes, you have all your carousel images ready. So once you have your pictures ready, take them into your editing software, add the hook and value text into each slide that Claude built, and you have a full carousel ready to post.

05:33What you're seeing on the screen right now has been made using one skill and one prompt. A year ago, this would have cost me hundreds of dollars to outsource. Now I just activate the skill and it's done.

05:44Let me show you this in real time. I'm going to take an actual line from the script of the video you're watching right now and turn it into a cinematic animation in under three minutes. So first, just activate the scale by typing this.

05:57Then I take this exact line from the video script, drop it in with this prompt, and hit send. And we get a complete ready to use prompt. Now I select Opus 4.8, drop this exact prompt in, and wait couple of minutes.

06:11You can see that it's actually using the motion graphic animator skill as it builds the animation, and we get this. I also took a few more lines from my script and had cloth turn each of them into its own animation. They all came out great, but this is the one I liked the most.

06:26Since it's an HTML file, you've got two options. Download it and run it through an HTML to m p four converter or just screen record the whole thing. Now if you want to get the same animation for vertical format, all you need to do is just use this prompt instead.

06:41Drop your script in, get the prompt back, select Opus 4.8, and a couple of minutes later, you get this. Honestly, it's wild that you can take any part of your script, give it the cloth, and get animations that look like this. And both of these formats took me less than five minutes.

06:55If you create content, you know how painful it is to write good scripts. So let me show you the exact prompt system I used to go from a content idea to a fully written short form video script in minutes. We already have the hard part done.

07:09Verlo gave us a proven content idea. Now we just feed that into cloth and let it do the rest. First, let's activate the script writing skill.

07:17Then I take the content idea from Verlo and turn it into a scroll stopping hooks using this exact prompt. Just insert the content idea into the brackets right here, and we get 10 scroll stopping hooks built around a topic we already know works. I just pick the strongest one and turn it into a full script.

07:35And if you want to change the hook a little bit, you can just ask Claude. Then I use this prompt and I insert the chosen hook into the brackets right here. And just like that, we get a finished script.

07:45Hook story, call to action, everything is ready to record. So we took one viral content idea that we got using Verlo. We turned that data into a carousel, and now we also have a full video script ready to record.

07:58And if you want to see how you can edit your videos faster using other cloth skills that I built, you can find that in this video right here. I'm building more of these skills and sharing them with you every week, so make sure to subscribe to not miss that. Thank you so much for watching, and I'll see you in the next

The Hook

The bait, then the rug-pull.

A creator opens by showing off thumbnails, motion graphics, and an Instagram carousel he says Claude built for him in minutes — then gives away the exact Skills behind all of it for free.

Frameworks

Named ideas worth stealing.

02:53concept

Idea-to-asset pipeline

Find a proven idea via Virlo trend data
Feed idea into a Claude Skill to get a structured prompt
Render the asset in GPT-image-2 (or Claude for animations)
Iterate the prompt with Claude until the result matches the vision

The repeated four-step loop used across every asset type in the video (thumbnails, carousels, animations, scripts): trend data in, Claude drafts the prompt, an external renderer produces the asset, Claude refines it.

Steal forany content-ops workflow that needs to mass-produce visual assets from a small number of proven ideas

CTA Breakdown