How to clone any Pinterest aesthetic with Claude

What you are actually building

The workflow is four moves, and each one hands its output to the next as a real file on your machine. That matters more than it sounds. Because every stage writes a file, you can stop halfway, fix the bit that drifted, and pick up again without re-running the whole thing.

Give Claude a browser and a folder

Claude can already see images and write prompts on any plan, including Free. The one thing it needs help with is the Pinterest part, because that means clicking around a live website. There are two honest ways to hand it a browser, and they suit different people.

Route B is what the rest of this guide assumes, because the file outputs are the part that makes this reusable. Here is the whole setup.

1. 💻 Install Claude Code. Follow Anthropic's setup guide. It is a terminal app that comes with a paid Claude plan, so there is no second subscription. When it is installed, make a new folder for this project (something like moodboard/), open your terminal inside it, and type claude.
2. 🌐 Add the browser. In that same terminal, run claude mcp add playwright npx @playwright/mcp@latest. That installs Playwright MCP, Microsoft's official browser tool for AI agents. Type /mcp inside Claude Code to check it shows as connected.
3. 🔐 Log in once, properly. The first time Claude opens the browser, ask it to go to pinterest.com and log in yourself in that window. Playwright keeps a persistent profile by default, so your session sticks around and you never do this again. Without it, Pinterest shows a login wall after a few scrolls and the hunt comes back thin.
4. 🖼️ Drop your reference in. Save the one image you want to clone into the project folder as reference.jpg. A screenshot is fine. It just needs to be the look you want, not the subject you want.
5. 🎨 Connect an image model. Pick one from the render options further down and connect it before you get to stage four, so the last step runs hands-off instead of stalling.

Read the style DNA

This is the prompt that does the heavy lifting, and it is the one most people write too vaguely. "Describe this image" gets you a caption. What you want is a spec sheet: named lighting, hex codes, lens behaviour, materials. Attach reference.jpg and paste this.

Look at reference.jpg and pull out its visual DNA in detail, as a spec another person could shoot from.

Cover: the lighting (direction, hardness, colour temperature, where the shadows fall), the colour palette as actual hex codes with a rough percentage for each, the composition and framing (crop, subject placement, negative space, apparent focal length and depth of field), the surface textures and materials, any post-processing character like grain or colour grading, and the mood in exactly three words.

Then write six Pinterest search phrases that would surface more images in this exact style. Write them the way real people tag things on Pinterest, so short aesthetic phrases rather than camera jargon, and make each one attack a different angle: the mood, the palette, the setting, the lighting, the era or aesthetic name, and the subject.

Save all of it to style-dna.md in this folder.

Send it hunting on Pinterest

Now Claude opens the browser and works through those phrases. Pinterest's search URL takes a plain query, so this is genuinely simple navigation rather than anything clever: pinterest.com/search/pins/?q=your+phrase.

Using the browser, open Pinterest and run each of the six search phrases from style-dna.md, one at a time, using the URL pattern pinterest.com/search/pins/?q=the+phrase.

For each search, look at the results properly and shortlist only the pins that match the lighting, palette and mood in the style DNA. Ignore anything that matches the subject but not the treatment, because the treatment is the thing we are cloning.

Across all six searches, pick the five closest matches overall, save each pin's image into a folder called refs/ as 01 through 05, and write refs/notes.md listing each file with its Pinterest URL and one line on what that pin adds to the style that the others do not.

Do not follow, save, comment on or send anything while you are in there. You are looking, not interacting.

Turn the pins into prompts

Here is the part most people miss, and it is the actual asset. The images are disposable. The prompts are the thing you keep, because a good prompt regenerates that style on any subject, on any model, next month and next year. This is where your moodboard stops being a folder and starts being a system.

Now write one image-generation prompt per reference in refs/, five in total.

Each prompt must be a single paragraph of 60 to 100 words that a text-to-image model can use with no other context, and it must carry the style DNA explicitly: the lighting setup, the exact palette (name the colours, do not say "warm tones"), the framing and depth of field, the surface textures, and the three-word mood.

Keep the treatment identical to the reference but swap the subject to: [describe your subject, for example "my amber skincare bottle" or "a workspace flat lay with a notebook and a coffee"]. End every prompt with the aspect ratio [4:5].

Save them to prompts.md, numbered 1 to 5, with the source pin URL and a one-line note under each explaining which part of the style that prompt is carrying.

Pick who renders it

Any image model can take these prompts. What changes is whether Claude can drive it for you or whether you paste the prompts in yourself. Both are fine. Hands-off is nicer, paste-in is cheaper to start.

Adding any of the hands-off options is the same three clicks. In the Claude app, open Customize, then Connectors, then Add custom connector, and paste the server address. Custom connectors work on every plan including Free, though Free is capped at one of them (Anthropic's connector docs). In Claude Code you add the same thing from the terminal with claude mcp add.

Take prompts.md and generate one image for each of the five prompts with my connected image tool, one at a time so nothing gets rate limited.

Save every result into a folder called out/, named 01 through 05 so each output matches its reference number.

After each generation, put the result next to its reference in refs/ and tell me in one line where it drifted: lighting, palette, framing or texture. If it missed on any of those, rewrite that prompt to correct the specific miss, note what you changed in prompts.md, and generate it once more. Two attempts per prompt maximum, then move on and flag it for me.

Save the whole thing as one skill

Doing this once is a nice afternoon. Doing it in one line every time you find a look you like is a system. A skill is a small folder with a SKILL.md inside that teaches Claude a process once, and it triggers automatically when your request matches. Paste this and Claude builds it for you.

Build me a reusable Claude skill called moodboard-clone that runs this entire workflow from a single reference image, and use skill-creator to package it.

The description should say: use this when I drop in a reference image and ask for a moodboard, a style clone or on-brand visuals.

The instructions are the four stages we just ran. One, read the image and write style-dna.md covering lighting, hex palette, composition, texture, mood in three words, and six Pinterest search phrases written the way people actually tag moodboards. Two, search Pinterest in the browser for each phrase and save the five closest pins into refs/ with their URLs in refs/notes.md. Three, write one 60 to 100 word generation prompt per reference into prompts.md with the source URL under each. Four, generate each prompt into out/ and compare every result back to its reference, retrying a maximum of twice.

Before starting, always ask me two things: the subject to swap in, and the aspect ratio. Never follow, save, comment or send anything on Pinterest.

1. 📦 Take the folder it hands you. skill-creator finishes with a skill folder containing a SKILL.md. Ask it to zip that folder so it is ready to upload.
2. ➕ Install it. In the Claude app: Customize, then Skills, then Add, then Upload a skill, and choose the zip. In Claude Code, drop the unzipped folder into ~/.claude/skills/. Skills work on every plan, Free included (Anthropic's skills guide).
3. 🔁 Run it on the next look you love. Drop an image in and say "moodboard this". It asks for your subject and aspect ratio, then works through all four stages on its own.

The four things that go wrong

Proof

Claude Code, official setup docs
Playwright MCP, Microsoft's official browser tool for agents
Getting started with Claude in Chrome
About custom connectors (remote MCP)
Creating custom skills