← sources

How are people actually getting Claude to build beautiful UIs instead of generic AI slop?

source: reddit (reddit:ClaudeAI)published 2026-09-28 09:11ingested 2026-09-29 02:38

retained text

I’ve been building websites and I keep running into the same problem.  
Claude is actually pretty good at structure, UX flow, information architecture, and getting a functional dashboard together.  
But when it comes to the **visual layer**, everything starts looking like generic AI-generated SaaS:  
cards everywhere  
random gradients / glows  
huge border radiuses  
Lucide icons everywhere  
pills and badges for no reason  
generic spacing  
generic dashboards that all look the same  
Basically… AI slop.  
What confuses me is that I keep seeing people on X, YouTube, etc. showing genuinely beautiful interfaces supposedly built with Claude/Opus.  
So I’m trying to understand what the **actual workflow** is behind those results.  
Are you guys really just giving Claude a good prompt?  
Or are you using things like:  
Figma references  
screenshots of existing designs  
Mobbin / Refero / Dribbble references  
shadcn / Radix / specific component libraries  
Claude Code + Playwright  
screenshot → vision critique → fix → screenshot loops  
image generation tools for custom illustrations/assets  
MCP/browser tools  
another model specifically for visual QA  
What I *really* want is something closer to this:  
Claude builds the page → opens the app → looks at the actual rendered UI → realizes that an illustration/icon/banner/etc. would improve it → creates or finds the asset → places it → takes another screenshot → critiques the result → iterates until the UI actually looks polished.  
Not just:  
“Make this dashboard more modern and premium.”  
…and then getting the same 4 cards + gradient + purple glow again.  
For people who are consistently getting **high-end, production-quality UI from Claude**, what does your actual workflow look like?  
Especially interested in hearing from people using Claude Code / Opus.  
Prompt examples, tool stacks, MCPs, visual feedback loops, or screenshots of your workflow would be hugely appreciated.  
I’m not really looking for another “use better prompts” answer. I’m trying to understand the repeatable process behind the genuinely good results.

document fetched 2026-09-29 00:23 · canonical https://reddit.com/r/ClaudeAI/comments/1ws9oix/how_are_people_actually_getting_claude_to_build