How I Used Gemini's Nano Banana to Generate Claude-Style Cover Images

When building claudefan.dev, I knew generic stock photos wouldn’t cut it. No glowing blue brains. No binary code tunnels. I wanted something that felt modern, approachable, and slightly hand-crafted — similar to the sketchy, hand-drawn aesthetic you see on Claude’s official website.

Since this blog is dedicated to building with Claude, why not match the visual identity?

I turned to Gemini’s Nano Banana, Google’s image generation model, to reverse-engineer Claude’s illustration style and generate consistent cover images. This post walks through my exact workflow — from analyzing the reference to a reusable prompt template you can steal for your own projects.

The Problem: Finding a Consistent Visual Identity

My first attempts leaned into trendy styles like Glassmorphism — sleek, frosted-glass banners with gradients and floating cards.

Prompt: “generate a blog cover image in Glassmorphism style”

The results were technically impressive but felt wrong. Too polished. Too corporate SaaS. They lacked the “human-in-the-loop” feel I wanted for a blog about working with AI.

I needed to pivot toward something more authentic.

The Breakthrough: Reverse Engineering Claude’s Style

I knew exactly what I wanted — the hand-drawn, whiteboard-doodle look from Claude’s website — but I didn’t know how to describe it to an AI image generator.

The breakthrough came when I stopped prompting for generation and started prompting for analysis.

I grabbed a reference image from Claude’s website and asked Gemini to explain it:

Prompt: “[Image from Claude’s website] explain the look and feel of this image. what style it is in. keep it short”

Gemini identified it as “Hand-Drawn Flat” or “Sketchy Minimalism” and extracted the key elements:

Element	Description
Mood	Casual, approachable, playful
Technique	Imperfect, organic “doodle” illustrations
Linework	Rough, uneven black lines (marker-like)
Palette	Matte colors — slate blue, terracotta, olive green

This was the turning point. I now had the vocabulary to describe exactly what I wanted.

Testing: Can Gemini Recreate Claude’s Style?

With this new vocabulary, I asked Gemini to create a cover image in the same style:

Prompt: “create a cover image in the style of Claude’s website — hand-drawn, sketchy minimalism, matte colored background”

It worked. Gemini produced a matte-colored card with rough marker-style illustrations — technical but human. The same aesthetic as Claude’s website, generated on demand.

The Master Prompt Template

One successful image isn’t a system. I needed something repeatable.

I asked Gemini to codify its own success:

Prompt: “give me this image style and theme as a prompt template, so I can reuse it”

Here’s the template it generated:

Sketchy Minimalism Master Prompt

A minimalist, hand-drawn illustration of [YOUR SUBJECT HERE] in the style of a rough marker sketch. Use thick, organic, uneven black lines on a solid, flat matte background (use colors like terracotta, slate blue, or olive green). The aesthetic should be casual, playful, and human-centric, resembling a quick napkin doodle or whiteboard brainstorming session. No gradients, no 3D effects, no shadows, high contrast.

This template decouples subject from style. I only change the bracketed text to generate consistent imagery for any topic.

Example: Generating a Cover Image

For my intro blog post, I needed to illustrate “developer + AI collaboration.”

Final prompt:

A minimalist, hand-drawn illustration of a pen, an AI brain, and a website in the style of a rough marker sketch. Use thick, organic, uneven black lines on a solid, flat matte background (use colors like terracotta or olive green). The aesthetic should be casual, playful, and human-centric, resembling a quick napkin doodle or whiteboard brainstorming session. No gradients, no 3D effects, no shadows, high contrast.

The result matched the exact visual language I’d established — consistent line weights, matte background, napkin-sketch charm.

The Workflow (TL;DR)

Here’s the repeatable process:

Step	Action
1. Find a reference	I used Claude’s website illustrations as inspiration
2. Analyze it	Ask AI to describe the style in detail
3. Extract keywords	Note mood, technique, palette, linework
4. Create a template	Ask AI to turn the style into a reusable prompt
5. Generate	Swap subjects, keep the style consistent

Key Prompts to Steal

For analyzing a reference image:

[Image Attachment] explain the look and feel of this image. what style it is in. keep it short

For creating a template:

give me this image style and theme as a prompt template, so I can reuse it

For generating images:

A minimalist, hand-drawn illustration of [SUBJECT] in the style of a rough marker sketch. Use thick, organic, uneven black lines on a solid, flat matte background (use colors like terracotta, slate blue, or olive green). The aesthetic should be casual, playful, and human-centric, resembling a quick napkin doodle. No gradients, no 3D effects, no shadows, high contrast.

Conclusion

Generating consistent cover art isn’t about typing wishes into a prompt box. It’s a workflow:

1. Explore — Try styles, be willing to discard them
2. Analyze — Use AI’s vision to dissect references and extract vocabulary
3. Templatize — Turn success into a reusable Master Prompt
4. Generate — Swap subjects, maintain consistency

By reverse-engineering Claude’s illustration style with Gemini, every post on claudefan.dev now matches the visual identity of the tool it’s about — without spending hours searching for stock art or hiring a designer.

---

Have a style you’ve templatized? Reach out on X @smhnkmrtweet and share your approach!