How to Create Professional Social Carousels Just by Talking to Claude
By Uramaki Studio Editorial Team
Describe the carousel you want in a chat, and it appears in your account — in your brand, with your fonts, ready to export. Here is exactly how it works, what it costs you in setup, and where it stops.
What this actually looks like
You open a chat with Claude and type one sentence: build me a five-slide carousel about why most skincare routines fail in winter, for women 25 to 40, using my brand.
A few seconds later the campaign exists. Not as five paragraphs in the chat window that you now have to rebuild somewhere — as a real carousel in your Uramaki account, with a headline and body copy on every slide, an image brief for each one, and your brand already applied.
You read it back. Slide three is vague, so you say so. It gets rewritten. You ask for the visuals, they render. Then you open the campaign, check how it looks, and export.
No copy-pasting. No retyping the same headline into a design tool. No version of the post living in three places at once.
What you can ask for
Everything below is a real thing you can say in the chat. You never learn commands or menus — you describe what you want in your own words.
| You say | What happens |
|---|---|
| "Build a 5-slide carousel about winter skincare mistakes" | A full campaign is created, planned slide by slide |
| "Slide 2 is too vague — give me a real number" | That slide is rewritten in your campaign |
| "Make all the headlines shorter, they're wrapping" | Every headline is tightened at once |
| "Now make the images" | The visual for each slide is generated |
| "Which brands do I have set up?" | Your brand kits are listed back to you |
| "Add a rule: never mention competitors" | Your brand guide is updated, and every future post obeys it |
| "What was I working on last week?" | Your recent campaigns come back with links |
The pattern worth noticing: these are not requests for text about your work. They are changes to your work.
A real example, start to finish
Here is a first message that works well:
Using my Acme brand, create a 5-slide Instagram carousel explaining why most skincare routines fail in winter. Audience is women 25–40 who already buy mid-range products. End on a soft call to action for the routine builder.
Six things are in that sentence: the brand, the number of slides, the platform, the topic, who it is for, and how it should end. That is roughly the minimum for something you will not immediately rewrite. Drop the audience and you get copy addressed to nobody in particular — the most common reason AI copy feels flat.
What comes back is not five interchangeable blocks. A five-slide carousel is built as a story:
- Slide 1 — a hook that goes against what people expect
- Slide 2 — why it matters right now
- Slide 3 — the method, concretely
- Slide 4 — proof tied to that method
- Slide 5 — the benefit, then the call to action
Ask for three slides and the arc compresses; ask for eight and it expands. The shape adapts to the length instead of padding it.
Why the copy comes back so short
The first reaction is often that the slides look thin. They are meant to. A carousel slide gets read in under two seconds, at arm's length, on a phone, usually one-handed.
So the copy aims for roughly:
- Headlines around 30 characters — short enough to hold one line at a readable size
- Body copy around 17 words per slide
- Every slide except the last ending on an open loop, so the swipe feels necessary
- The last slide being the call to action and nothing else
That last one is the rule people break most often. A summary slide right before the ask feels thorough, and it drains the ask. If you only change one habit, change that one.
These are targets, not hard cuts — nothing gets chopped mid-sentence. But copy that reads beautifully in a document is almost always too long on a slide.
If your openers are the weak part, this breakdown of hooks that stop the scroll pairs directly with this.
Making it look like your brand, not like AI
Speed on its own just produces bad carousels faster. What separates output that looks like a brand from output that looks like a prompt is almost entirely set up once, in your brand kit — then inherited by everything after.
- Fix your fonts once. Nothing reads as machine-made faster than typography that changes between posts.
- Write your rules as bans, not wishes. "Never use exclamation marks" gets obeyed. "Be professional" does not mean anything to anyone.
- Be specific about images. "Product on wet stone, cold morning light, shallow focus" gives you a brand. "Nice product photo" gives you stock.
- Keep the final slide down to the ask alone.
Because your tone of voice and your rules are attached to every generation, you state them once rather than re-explaining them in every conversation. That is the difference between a shortcut and a system.
One small thing that trips people up: on layouts where the copy sits beside the image in a narrow column, the text is written as flowing sentences rather than bullet points. Bullets in a narrow column read like a fragment of a slide deck. If you want a list, pick a layout built for one.
Not set up yet? The visual brand identity guide is worth twenty minutes before you automate anything.
Where the chat stops
Worth being straight about this, because it decides how you plan your time.
One request gets you finished slides: the assistant plans the carousel, generates the pictures, puts your copy, logo and swipe label on them, and hands back a download link per slide. What it will not do well is fine-tuning — nudging a headline up, trying a different layout on slide three, changing one word and seeing it land.
So the split is not between planning and producing any more. It is between producing and adjusting. The chat takes the part that benefits from talking: thinking, structuring, writing, generating, rendering. The app takes the part that benefits from seeing and moving things — which is why every answer still ends with a link to the campaign.
The useful test is not whether everything happens in one window. It is whether you ever type the same sentence twice.
Setting it up (about a minute)
There is nothing to install and no code. You are copying one link into Claude, then signing in the way you sign in to anything else.
- In Claude, open Settings, then Connectors.
- Choose Add custom connector.
- Paste the link below and confirm.
- Uramaki opens in a new tab: sign in and approve.
- Back in Claude, you can start asking.
https://uramakistudio.com/api/mcpThat approval screen matters more than it looks. It shows exactly what the connection will be allowed to do, and it only ever covers your own account. Your password never reaches Claude — you sign in on Uramaki, and Claude gets a key it can use, which you can take back at any moment by removing the connector.
One thing to expect: Uramaki is not in Claude's built-in list of apps yet, so you add it yourself with the link above. Once it is in, it behaves exactly like the ones that are listed.
The same steps with screenshots and troubleshooting live on the connection page.
The technical name for all this: MCP
You do not need this section to use any of the above. It is here because you will see the acronym, and knowing what it means makes it less alarming.
MCP stands for Model Context Protocol. It is an open standard — not owned by any one company — for letting an AI assistant use an outside app. The app publishes a list of things it can do; the assistant reads that list and picks what it needs. That is genuinely the whole idea.
Two things it is not, because both are common worries:
- It is not the assistant clicking around your screen. Nothing is being remote-controlled, and nothing is being scraped.
- It is not a copy of your data sitting somewhere else. Your campaigns stay in your account; the connection reads and writes there, live, only when you ask.
The practical upshot for you is a change in what the assistant is. Without it, it is a very good writing partner that hands you a draft. With it, it does the work in the place where your work already lives.
Prefer a different assistant?
Because the standard is open, the same link works with any assistant that supports these connections. There is nothing Claude-specific about it.
A few apps only support connections that run on your own computer. If yours is one of them, this command bridges to the same thing — paste it where that app asks for a connection, and the first run opens the same sign-in page:
npx -y mcp-remote https://uramakistudio.com/api/mcpIf that line means nothing to you, you almost certainly do not need it.
If something does not work
| What you see | What it usually is |
|---|---|
| The connector will not add | The link is slightly off. It has to end in /api/mcp |
| The sign-in page does not load | You are not signed in to Uramaki in that browser. Sign in first, then try again |
| Claude says it only has permission to look, not change | Some permissions were skipped. Remove the connector, add it again, approve everything |
| Claude writes the carousel in the chat instead of making it | It did not reach for the connection. Say "create it in Uramaki" and it will |
| The slides are there but the images are not | Images are a separate step. Ask for them once the words are right |
Who this is actually for
If you post twice a month, setting this up will cost you more than it gives back. That is a fair trade to decline.
It pays off in three situations:
- Volume — several carousels a week, where the bottleneck is making them, not thinking of them.
- More than one brand — agencies and freelancers switching between clients, where consistency is the first thing to slip.
- Thinking out loud — if you find your angle by talking it through, and would rather the post exist by the end of that conversation than start it afterwards.
If you would rather see the app side without any of the chat, here is how Uramaki works on a real example.
FAQ
Do I need to be technical to set this up?
No. You paste one link into a settings screen in Claude, sign in to Uramaki in the tab that opens, and approve. There is nothing to install, no code, and no API key to look after.
Can Claude see everything in my account?
It can read and change the campaigns and brand kits in your own account, which is the point. It cannot see other people's work, your password, your payment details, or your social logins. Removing the connector ends its access immediately.
Will the finished images appear in the chat?
Yes. The assistant renders the slides with your copy, logo and swipe label already on them, and gives you a download link for each. You still get a link to the campaign in Uramaki, which is where you go to adjust a layout or rewrite a line and see the result.
How many slides should a carousel have?
Five is a solid default: enough room for a hook, a reason, a method, proof and an ask, without requesting more swipes than most people will give you.
Does this work with other AI assistants?
Yes. The standard is open, so the same link works with any assistant that supports these connections. Apps that only support connections running on your own computer can bridge to it.
Is Uramaki an official Claude app?
Not yet — you add it yourself as a custom connector, which takes about a minute. Once added it works exactly like the built-in ones.
Ready to generate faster campaigns?
Generate your first campaign free on Uramaki Studio.