The best Claude Design workflow you’ll ever see…
TL;DR
Alex Finn’s workflow is: explore in ChatGPT, refine in Claude Design — he uses OpenAI’s new image model for cheap, high-volume logo and brand experimentation, then hands the winning direction to Anthropic’s Claude Design for a near one-shot design system and prototype.
The bottleneck is Claude Design’s low usage limits — Finn says many people burn through Anthropic’s quota after just 2-3 design attempts, so doing the messy trial-and-error in ChatGPT helps avoid wasting Claude turns.
He starts with visual inspiration from X, not a blank prompt — in the demo for “Henry Intelligent Machines,” he grabs 3-5 references like gradients, the Orlando Magic logo, and the Nous Research logo, then asks ChatGPT to generate sheets of octopus- and owl-themed logo concepts.
The key move is iterative narrowing, not chasing a perfect first output — Finn repeatedly screenshots promising options, asks for flatter and more minimal versions, then refines gradients and colorways over 5-6 rounds until he lands on an octopus app-icon logo.
ChatGPT does more than images here — it also writes the Claude prompt — after generating a branding guide, logo variants, merch, typography, and colors, he asks ChatGPT to summarize the entire brand into a prompt for Claude Design.
Claude Design shines once the direction is clear — after uploading the app code, the chosen logo sheet, and the ChatGPT-generated prompt, Finn gets a full design system in about five minutes and then a high-fidelity dashboard prototype he says can be handed directly to Claude Code.
The Breakdown
Two best-in-class tools, two annoying weaknesses
Finn opens with a bold claim: combine Claude Design and ChatGPT’s new image generation model, and even “zero design or artistic skills” people can make beautiful apps. His framing is practical, not hypey-for-hype’s-sake: Claude Design is amazing but gets kneecapped by low usage limits, while ChatGPT image gen is cheap, high-limit, and excellent at visuals but not full design systems.
Stage 1: Hunt for inspiration on X
Instead of prompting from scratch, he scrolls X and saves whatever catches his eye into a visual inspiration folder — gradients, logos, clean shapes, anything with the right vibe. In the demo, that includes the Orlando Magic logo and Nous Research branding, plus his own preferences like octopi and owls for an app called Henry Intelligent Machines.
Stage 2: Use ChatGPT to find a direction, not the final answer
He prompts ChatGPT with the app description — an AI agent that autonomously builds revenue-generating businesses — and explicitly tells it to “use your new image gen model,” because otherwise it may wander into weird outputs like PDFs. His metaphor here is memorable: you’re “in the middle of the forest” and just looking for a direction, not perfection.
Stage 3: Screenshot, simplify, and keep narrowing
Once a few logos stand out, he screenshots them and asks for flatter, more minimal versions with many variations. He recommends doing this refinement loop 5-6 times: pick a couple, expand on them, then repeat until the identity sharpens.
Add one twist: refine while still letting the model surprise you
Finn doesn’t overconstrain the model too early. After narrowing to four options, he asks ChatGPT to expand on them but also make some “out of the box” versions, which leads him toward a futuristic octopus direction and a gradient style he says he hasn’t really seen elsewhere.
From logo to full brand pack
After settling on an octopus-style app icon and a preferred top-row gradient, he asks ChatGPT to create a branding guide and even some merch. The result includes logo variations, typography, brand values, a credit card mockup, and merch — fun extras, but also a compact reference pack for the next tool.
Let AI write the prompt for Claude Design
One of Finn’s strongest practical points is that “AI is the best tool in the world for writing prompts.” So instead of manually summarizing the brand, he asks ChatGPT to generate a prompt describing the logo, colors, and overall direction for Claude Design.
Claude Design builds the system, then the prototype
Inside Claude Design, he uploads the prompt, the image assets, and even the app’s codebase so the tool can understand features and functionality. About five minutes later, Claude generates a complete design system — colors, spacing, typography, components, gradients — then uses it to create a high-fidelity dashboard prototype that Finn says can be shared directly with Claude Code to implement, taking you from “AI slop” to a polished app UI.