I Gave Claude Design a Second Chance: How Good Is Its New Design System, Really?

24 May 2026 02:37 18,800 views
This hands-on review puts Claude Design’s new design system features to the test on a real clothing brand. The results show impressive potential—but also major limits around accuracy, usability, and strict credit caps.

Claude Design promises something ambitious: drop in your brand assets once, and the AI will generate on-brand websites, flyers, and more with just a prompt. After an underwhelming first try cut short by strict usage limits, this follow-up test focuses on the feature that might matter most—Claude’s design system.

Setting Up a Design System in Claude

The test brand here is Paint Factory, a small clothing label built around one-of-one, hand-painted pieces. It’s a playful, relatively loose brand—no 40-page style guide, just a logo, a general visual vibe, and some basic rules like a white dotted background, Poppins as the default font, and bold uppercase titles.

That makes it a good candidate for Claude Design’s system feature, which works a bit like a smarter version of Canva’s Brand Kit. You feed it:

• Brand description and SEO copy
• Logos and illustration files (like an Illustrator logo file)
• Website exports (such as .webarchive versions of the home and about pages)
• Supporting graphics and squiggly assets used across the brand
• Notes on tone, layout, and typography

Claude then processes everything and generates a design system in a markdown file. This is similar to Google’s Design MD approach—your brand is encoded in a structured, machine-readable format so the AI can reuse it consistently across new designs.

What Claude Actually Built

Once the system finished processing (it took longer than the claimed five minutes), the first surprise was big: Claude had quietly built a fully functional website layout for the brand.

The site mirrored the existing Paint Factory site structure—hero section, product grid, and overall flow—but with Claude’s own interpretation of the visuals. The layout looked clean, modern, and surprisingly polished, with defined spacing, corner radii, shadows, and reusable components like grid cards.

However, there were immediate issues:

• Colors: The palette didn’t match the real brand colors. Claude hadn’t pulled the exact values from the assets, so they had to be corrected manually.
• Borders and style details: Some cards used partial borders and softer styling, while the brand leans more toward bold, comic-style outlines. These needed tweaking to feel authentic.

On the positive side, the underlying structure of the system was strong. You could drill into components, adjust spacing, borders, and shadows, and see how the system was thinking about the brand. For a first release, that’s impressive—and it lines up with what we’ve already seen in deeper breakdowns like Claude Design: Anthropic’s New UI Superpower Explained.

Testing Claude Design With a Flyer

To see how well the design system actually works in practice, the next test was a flyer. The idea: plug in the Paint Factory system, give Claude a detailed prompt with event info and messaging, and see if it can produce a usable, on-brand layout with lots of content.

Claude’s interface splits into two sides: a chat-style prompt area on the left and a live design canvas on the right. You can attach your design system as context, then ask it to “Make a flyer with this information” and paste in your copy.

When the flyer finished generating, the first impression was surprisingly good:

• Visually, it looked like a real, professionally designed flyer.
• The overall style felt close to the Paint Factory brand—bold, playful, and graphic-heavy.
• The layout was responsive on the canvas, stretching and compressing as you zoomed, which is unusual for a static flyer but shows how web-first the tool is.

But a closer look revealed problems that make it hard to use straight out of the box:

• Layout glitches: Some text overlapped or ran off the canvas, like a section where a line of copy was covered by a decorative squiggle.
• Color inconsistency: Multiple shades of red appeared, and the yellow didn’t match the brand yellow at all.
• Off-tone copy choices: A section titled “The Rules” clashed with the brand’s emphasis on creative freedom and individuality, even though the prompt clearly leaned into self-expression.

The result felt like a strong first draft from a junior designer: promising, stylish, and close to the mark—but not something you could print or publish without a lot of manual cleanup.

Where Claude Design Shines (and Where It Breaks)

Across both the website and flyer tests, a pattern emerges.

What Claude Design does well:

• Generates complete, coherent layouts (web pages, flyers) from brand context and a prompt.
• Learns enough of the brand’s visual language to feel “in the ballpark,” especially for younger or less rigid brands.
• Encodes the brand in a reusable design system, which is a powerful foundation for future generations.

Where it struggles:

• Precision: Exact colors, spacing, and fine layout details often need correction.
• Strict brand guidelines: The more specific and established your brand is, the more the small inaccuracies will bother you.
• Editing: Fixing local issues (like removing a single squiggly line or adjusting a specific overlap) still feels clunky and slow.

Most importantly, the tool is clearly not yet at the “no designer needed” stage. It’s closer to a fast idea generator and layout assistant than a final-output machine. That aligns with the broader takeaway from earlier testing: Claude Design is not the end of design—but it’s not ready yet.

The Biggest Dealbreaker: Usage Limits

Even if you’re willing to tweak layouts yourself, there’s a practical problem: credits. In this test, a $20 Claude subscription translated into roughly 20 minutes of heavy design usage before hitting a hard cap.

After building the design system, generating a website, and creating a single flyer with a few attempted edits, usage hit 96%. Once you’re out, you’re done until the weekly reset—no top-up, no extra headroom. That makes it nearly impossible to iterate the way real design work requires.

Design is inherently iterative. You rarely get it right on the first try. You need room to:

• Try multiple variations
• Refine copy and hierarchy
• Fix visual bugs and spacing issues
• Explore alternative directions

Claude Design’s current credit system cuts that process short, turning what could be a powerful workflow into a frustrating race against the meter.

What This Means for Designers and Non-Designers

Right now, Claude Design seems best suited for people who don’t have strong design skills but need decent-looking interfaces or marketing materials—especially developers already using Claude for code. For them, it’s a huge upgrade over starting from a blank canvas.

For professional designers, it’s more of an experimental tool than a production-ready solution. It can help with:

• Rapid concept exploration
• Generating starting points for layouts
• Translating loose brand ideas into something visual

But it doesn’t replace the core of design work, which is less about clicking tools and more about:

• Understanding what makes good, timeless design
• Knowing your audience and how to speak to them visually
• Making strategic choices about hierarchy, tone, and brand expression

AI can generate endless variations, but it still needs a human art director to decide what’s right, what’s off, and what truly fits the brand.

In its current form, Claude Design is exciting, promising, and occasionally impressive—but held back by strict credit limits and rough edges that make it hard to rely on for serious, iterative work. If those constraints ease and the editing experience improves, it could become a genuinely powerful part of the modern design toolkit.

Share:

Comments

No comments yet. Be the first to share your thoughts!

More in Claude