Build Consistent Brand Images with Claude Skills [System Breakdown]

18 min readBy Johnathan Chen

Brian Casel spent a week building a complex N8N automation to generate consistent brand images. It failed. Then he rebuilt it as a Claude Code skill in 30 minutes—with better results. Here's the complete breakdown of his system for creating brand-consistent AI images that actually works.

TL;DR:

  • •

    The problem: AI image models are great at one-off images, but you need a repeatable system that maintains brand consistency across dozens of images without recrafting complex prompts each time.

  • •

    Failed approach: Brian spent a week building an N8N automation workflow with nodes, API calls, and conditional logic. It technically worked but produced garbage output that didn't match brand guidelines.

  • •

    Winning approach: Claude Code skill built in 30 minutes. Uses Claude Opus for intelligence + Google Gemini API for image generation. Maintains full brand guidelines, suggests concepts, and produces consistent results.

  • •

    The secret: Spend most time developing comprehensive brand visual guidelines (visual world, illustration aesthetic, idea-to-illustration mapping) before building any automation.

The Challenge: Brand Visual Consistency

Brian Casel wanted his Builder Methods brand to have a strong visual presence. Like the vibe you get when you walk through a well-designed storefront—everything cohesive, intentional, distinctive. His writing and videos set the tone, but visuals complete the experience.

The problem? He's not an illustrator. And he doesn't need just one great image—he needs dozens. Every page, every piece of content, every thumbnail needs to follow the same visual aesthetic.

AI image generation has gotten incredibly good at creating one-off images. But what Brian needed was a repeatable system that could maintain his brand aesthetic without him recrafting complex prompts every single time.

The Turning Point

After spending a week building a complex N8N automation that produced garbage output, Brian rebuilt the entire system as a Claude Code skill in 30 minutes. Not only was it faster to build—the output quality was dramatically better.

One-Off vs Repeatable Needs

This is the fundamental difference that most people miss when working with AI image generation:

One-Off Images

AI models excel at creating individual images from detailed prompts. You describe what you want, tweak the prompt, regenerate a few times, and get something great.

Problem: Not scalable. Every new image requires the same manual effort.

Repeatable Systems

A system that encodes your brand guidelines, suggests concepts based on content, and generates consistent images without recrafting prompts.

Solution: Build once, use forever. Consistent aesthetic across all content.

The key insight: You're not building an image generator—you're building a brand visual system. That's why Claude Code skills are perfect for this. The intelligence lives in Claude, while the image generation is just one tool it uses.

Step 1: Develop Brand Visual Guidelines with Claude

This is the most crucial step and where Brian spent the most time. Before building any automation, you need to define your visual brand identity. He did this through a long conversation with Claude in a dedicated project.

Three Core Artifacts

How Long This Takes

This phase took Brian several days of back-and-forth conversations with Claude. It's worth the investment—these guidelines become the foundation for everything. Don't rush this. The automation is easy; getting the brand definition right is the hard part.

Step 2: Prototype with AI Image Generation

Once you have visual guidelines, test them with an AI image model. Brian used Google Gemini (their image generation API) to prototype different styles.

Find Inspiration on Dribbble

Browse illustration styles on Dribbble. Paste images into Claude and have it describe the styles. Use these descriptions as starting points for your prompts.

Tip: Look for styles that are distinctive, flexible enough for various subjects, and not overly "AI-looking."

Test Multiple Variations

Experiment with different line weights, levels of detail, color approaches, and composition styles. Evaluate each for:

  • • Does it feel distinctive and branded?
  • • Is it flexible across different subjects?
  • • Does it avoid generic AI aesthetics?
  • • Will it remain consistent across generations?

Technical Note

AI image generation typically produces flat PNG images, not truly transparent backgrounds. You'll likely need to post-process images in Photoshop or similar tools to create transparent versions for your site.

The Overengineered N8N Workflow (What Didn't Work)

After establishing his guidelines, Brian spent a week building a complex N8N automation. The idea: use Slack as the interface, have everything happen automatically through nodes and workflows.

The N8N Architecture

1

Slack Webhook Trigger

Message in Slack channel fires webhook to start workflow

2

Load Guidelines & Variables

Pull in all visual guidelines, prompts, and brand assets

3

Switch & Conditional Logic

Route through different paths based on state (new request, gathering info, generating concepts, etc.)

4

Generate Concepts → User Selection

Present 3 concepts to user, wait for A/B/C selection

5

Image Generation API Call

Call Google Gemini API with assembled prompt

6

Save to Google Drive & Notify

Store generated image, send link back to Slack

Why It Failed

The workflow technically worked—it delivered images to Google Drive. But the output was way off from brand guidelines:

  • • Too much detail (busy, cluttered illustrations)
  • • Ignored aesthetic guidelines (wrong line weight, shading)
  • • Wrong subject matter (illustrated entire offices instead of single objects)
  • • Inconsistent style across generations

The fundamental problem: By breaking everything into discrete nodes and rigid logic, Brian stripped away the intelligence that makes AI useful. The model couldn't think—it could only execute predefined steps.

The Claude Code Skill Solution

After giving up on N8N, Brian remembered Claude Code skills. The key insight: Claude Code itself could be the tool for the job—not just a way to build an app, but the actual interface for the workflow.

Claude = Intelligence

Claude Opus reads guidelines, understands context, suggests concepts, and reasons about brand identity. The intelligence layer that was missing from N8N.

Gemini = Generation

A simple Python script calls Google Gemini's API for actual image generation. Claude assembles the prompt based on all guidelines.

Skill = System

The skill packages everything into a self-contained system: guidelines, sample images, prompts, mapping logic, and generation script.

What's Inside the Skill

skill.md

Core skill file that Claude Opus reads. Defines workflow: create project folder, generate concept options, wait for user choice, reference guidelines, invoke generation script.

Brand Guidelines Files

Visual world document, aesthetic style guide, idea-to-illustration mapping, brand colors—all the artifacts created in Step 1.

Sample Illustrations

Example images that demonstrate desired style. Help train Claude on what illustrations should look like.

generation_script.py

Python script (written by Claude) that calls Google Gemini API. Pulls in all references, assembles the prompt, handles API authentication.

Result: 30 Minutes to Build, Perfect Output

The skill took 30 minutes to build (with Claude's help generating all the code). Output quality immediately matched brand guidelines. No more fighting with nodes and logic—just describe what's needed, choose from concepts, get consistent results.

Complete Step by Step Tutorial

Here's how to build your own brand illustrator skill following Brian's system:

Step 1: Develop Visual Guidelines (Most Important)

1. Start a Claude Conversation

Create a dedicated Claude project for your brand. Tell Claude you want to develop a visual brand identity.

2. Engage in Back-and-Forth

Define your brand, target audience, what you want to resonate with. Let Claude ask questions and suggest ideas. Iterate until you're aligned.

3. Define Your Visual World

Document the subject matter—types of objects and scenes that would appear in your visuals. Focus on what, not how.

4. Create Idea-to-Illustration Mapping

Build a guide for how to analyze content and suggest illustration concepts. Map content themes to specific objects or scenes.

5. Establish Aesthetic Guidelines

Lock in exactly how illustrations should look: line weight, detail level, colors, lighting, texture, composition rules.

Step 2: Prototype with Image Generation

1. Use Google Gemini (or similar)

Test your style descriptions with an AI image model. Use the aesthetic guidelines Claude helped you write as prompts.

2. Experiment with Variations

Try different colors, levels of detail, line thickness. Evaluate if the style is flexible, distinctive, and not overly AI-looking.

3. Refine Guidelines

Based on what works, update your aesthetic guidelines. Iterate until you have a style you're happy with.

Step 3: Build the Claude Code Skill

1. Start New Claude Conversation for Skill Building

Provide Claude with your established visual guidelines. Tell Claude you want to build a skill.

2. Claude Suggests Structure

Claude will use its skill creator skill to suggest the structure and provide files to download.

3. Integrate into Claude Code

Place the skill in .claude/skills/brand_illustrator/

4. Include All References

Visual world doc, aesthetic style, prompts, mapping guide, brand colors, sample illustrations.

5. Python Script for API Integration

Claude writes the Python script to integrate with Google Gemini's API. Add your API key to .env file.

6. Define skill.md Workflow

Outline the workflow: create project folders, generate concept options, wait for user choice, reference guidelines, invoke generation script.

Step 4: Use the Skill

1. Invoke Skill in Claude Code

brand_illustrator

2. Describe What You Need

"I'm creating a blog post about systems thinking. I need a hero image."

3. Claude Gathers Info

Claude will ask for color preference, dimensions, any specific requirements.

4. Choose from Concepts

Claude presents 3 illustration concepts based on your content and mapping guide. Select A, B, or C.

5. Image Generated

Claude crafts the prompt, runs the Python script, generates the image. Saved to project folder.

6. Regenerate if Needed

You can ask Claude to regenerate with tweaks until you get exactly what you want.

Key Takeaways

1. Spend Time on Guidelines, Not Automation

The most important work is developing comprehensive brand visual guidelines. This took days. The automation took 30 minutes. Get the foundation right first.

2. Claude Code Skills > Complex Automation Tools

N8N stripped away AI intelligence by forcing everything into discrete nodes. Claude Code skills keep the intelligence layer while still automating the workflow. Result: better output, faster build time.

3. Repeatable Systems > One-Off Images

You're not building an image generator—you're building a brand visual system. The goal is consistency at scale without manual prompt crafting each time.

4. Claude = Intelligence, Other Models = Tools

Claude provides reasoning and context understanding. Google Gemini (or other models) handle specific tasks like image generation. Combine them for best results.

5. Claude Code Can Be the Application

You don't always need to build a separate app. Claude Code itself can be the interface for business workflows. Skills package everything into a self-contained system.

6. Three Core Artifacts Define Your Brand

Visual World (what to show), Idea-to-Illustration Mapping (how to choose), Aesthetic Guidelines (how it should look). Get these right and everything else follows.

Final Thoughts

Brian's experience teaches an important lesson: sometimes the most sophisticated solution is actually the simplest one. He overcomplicated the problem by assuming he needed complex automation tooling.

The real complexity was in defining the brand visual system—the guidelines, the aesthetic, the mapping logic. Once that was clear, the automation became trivial. Claude Code skills provided exactly the right level of intelligence and flexibility.

If you're building any repeatable AI workflow—not just image generation—consider whether Claude Code skills might be the better approach compared to traditional automation tools. The model can think, reason, and adapt. Don't strip that away by forcing everything into rigid logic gates.

Want to Learn More About Building with Claude Code?

Get weekly insights on AI automation, Claude Code workflows, and practical techniques for building smarter systems.

Subscribe to Builder Briefing