---

What this workflow does

You connect Claude Code to your browser through Playwright MCP. From that point on, Claude can operate any website the way you would: click, type, upload files.

The flow: you hand Claude a reference image and tell it what your post is about. Claude reads the style from the reference (composition, color values, lighting), builds a dedicated image prompt for each blog post and then takes over the browser: open Pollo AI, upload the reference image, paste the prompt, start the generation. One image per post, all in the look of your reference.

The tools you need

Tool	Role	Cost
Claude Code	Runs the whole workflow: writes the prompts and operates the browser	Paid (Claude subscription)
Playwright MCP	Gives Claude full control over your browser	Free
Pollo AI	The platform where the images get generated	Requires credits

Step 1: Check that Claude Code is installed

Open your terminal (Mac: ⌘ + space, type "Terminal". Windows: PowerShell) and run:

claude --version

If you see a version number, you are good to go. If not, install Claude Code first via claude.com/claude-code.

Step 2: Install Playwright MCP

The most important step, a single command:

claude mcp add playwright npx @playwright/mcp@latest

Hit enter, done. From this moment on, Claude can control your browser.

If you get an error, check Node.js first:

node --version

No version number? Grab the LTS release from nodejs.org, install it and repeat the command above.

Step 3: Sign in to Pollo AI

Open pollo.ai in your browser and log in before you start Claude. A login screen in the middle of the run interrupts the automation. Also check your balance, every generation costs credits.

Step 4: Start Claude Code and hand over the prompt

Start Claude Code in your terminal:

claude

Then give it the job. My prompt for a single blog image:

Analyze the reference image at [PATH TO IMAGE].
Extract composition, color values, style and lighting.

Then read my blog post [LINK OR FILE] and build an image prompt
that fits the topic and stays exactly in the style of the reference.

Then open Pollo AI, upload the reference image, paste the prompt
and start the generation.

Batch mode: one image per blog post

Once the single run works, you scale with the same pattern:

Analyze the reference image at [PATH TO IMAGE]:
composition, color values, style, lighting.

Create a dedicated image prompt in the style of the reference
for each of the following blog posts:

1. [TITLE OR LINK BLOG POST 1]
2. [TITLE OR LINK BLOG POST 2]
3. [TITLE OR LINK BLOG POST 3]

Then open Pollo AI and generate the images one after another:
upload the reference image, paste the prompt, start the generation.

What happens next

1. Claude analyzes the reference image and locks in the style as a rule set
2. Each blog post gets its own image prompt, on topic, visually identical to the reference
3. Playwright opens the browser and navigates to Pollo AI
4. Claude uploads the reference image, pastes the prompt and configures the settings
5. The generation starts, the finished images land in your Pollo media library

While Claude works: hands off the machine. Moving the mouse, typing or switching tabs can interrupt Playwright. Kick off the workflow, get a coffee, collect the results.

Common mistakes

• The magic sentence is missing. Without "Then open Pollo AI", Claude stays in the terminal and only gives you text.
• Not logged in. If Pollo AI shows a login screen, the automation breaks. Sign in first.
• Credits not checked. Before a batch run, check your balance.
• Swapping the reference mid-series. That kills the consistency you built the setup for. One reference per series.
• Vague topic input. "Make a nice image" produces something random. The more concrete the blog post context, the better the prompt lands.

Beyond blog images

The same setup works for anything built from a reference plus a prompt:

• UGC ads and social media posts in a consistent look
• Carousel slides that match your brand
• Product photos on new backgrounds
• Cinematic video spots: Pollo AI ships with Seedance 2.0. You let Claude describe scenes, camera moves and lighting and generate short clips from it, same workflow, different model. For story videos directly inside the Pollo interface: Cinematic Videos.

And if you want to go deeper into browser automation with Claude Code (screenshots, tests, full render pipelines): Claude Code + Playwright.