You Don't Need Claude Design. You Need These 3 Prompts.
Published: April 24, 2026 • 7 min read
Anthropic shipped Claude Design on April 17, 2026. One week ago.
I'm not surprised. I published the methodology behind it on November 25, 2025.
That's not a coincidence. That's the second time in five months that something I shipped has previewed Anthropic's roadmap. And I'm starting to think my brain is telepathically connected to the brains of Anthropic engineers.
Let me show you the receipts.
Quick context for anyone new here. From October 23 to December 13, 2025, I paused my job search for 50 days and called it the SDR Era, Skill Development and Refinement Era. Instead of sending more applications into the void, I focused on buiilding public proof of work. Projects. Case studies. Tools I needed myself. By the end of those 50 days, I had shipped 3 products, launched the Claude God Tips series, invented the LLM Instance Cloning concept, and written over 80 blog posts. Both of the projects in this post were born inside that sprint.
The Pattern (Two Examples Is Not a Coincidence)
Example one. Back in November 2025, I was trying to make images on my portfolio website look sophisticated using the Canva MCP via Claude, and the designs looked like a toddler had described a premium template to a stressed-out intern. The colors never respected my prompts. The layouts were mediocre. So I pivoted to something weird: I started taking screenshots of single HTML pages I'd build with Claude, and using those as cover images.
That pivot became a methodology I published on November 25, 2025: treat HTML pages as the design surface, screenshot them, ship the screenshot. Five months later, on April 17, 2026, Anthropic shipped Claude Design. Same primitive. HTML, CSS, and JavaScript as the design output format. Not Photoshop files. Not Figma exports. Code that renders as a design.
Example two. Last year, in that same era, I was deep in building ViteHero, an app meant to help non-developers generate sophisticated images using HTML, CSS, and JavaScript which has now been converted to an MCP(Model Context Protocol). While building it, I kept running into visual bugs on the front-end. And every time one showed up, I had to type out exactly what was broken or send screenshots so Claude could fix it. After enough rounds of that, I got lazy. I did not want to describe one more layout bug or send one more screenshot.
So I asked an obvious question: what if Claude could just see the app and fix it itself? The unlock was already in my hands. I was building screenshot-automation scripts inside ViteHero anyway. I pointed those scripts at my own front-end, fed the screenshots to Claude, and let it diagnose its own work.
I wrote that workflow up as a case study on December 1, 2025. Two and a half weeks later, on December 18, 2025, Anthropic launched Claude in Chrome to all paid plans. Same primitive. Same problem. Different surface area. I only found out about Claude in Chrome around late January this year.
Two examples. Same shape. Both times, I shipped first. I'll let you decide if that's telepathy or just paying attention to the same problems Anthropic engineers were paying attention to.
What I Actually Built (The Original Methodology)
Here's the short version of how I got there.
I started by trying to wire Canva up to Claude through an MCP integration. The output was bad. Generic. Off-brand. The kind of design that makes you close the tab.
So I tried something else. I noticed that when I deployed a new website to Vercel, the auto-generated preview screenshot looked stunning. Better than anything Canva was giving me. The realization landed hard: the website itself was the design.
So I flipped the workflow. Instead of generating a static image, I started telling Claude to build a single HTML page styled exactly the way I wanted, then take a screenshot of it. The screenshot became the cover image. The colors respected the prompt because the colors were literal CSS. The typography respected the prompt because the typography was literally a font import.
Once it worked, I wrote a slash command in Claude code called hero-generator. You pass a project name and a description. It outputs a polished image. Without writing a single design file.
That's the primitive. Treat HTML/CSS/JS as a design surface. Render it. Screenshot it. Done.
Here are four examples of what this methodology produces. Keep reading, I'll share the exact prompt that builds them.
What Claude Design Looks Like (And What I Can Honestly Say About It)
Claude Design ships HTML, CSS, and JavaScript artifacts. Same primitive I've been using for five months.
Anthropic hasn't disclosed the rendering pipeline, and I'm not going to pretend I've reverse-engineered it. I don't need to know how they built it. The artifact they ship is the same artifact I've been shipping.
What I can confirm from the announcement is that it's paid-plan only (Pro, Max, Team, Enterprise), and it's off-by-default for Enterprise.
I'll do a proper breakdown in a future post, once I've put it through real work. But the point of this post isn't to review Claude Design. The point is what I'm about to share with you next.
The Workaround (For Everyone Who Doesn't Have a Paid Plan)
Claude Design is gated behind a paid plan. My methodology is not.
I packaged the entire workflow into an AI Specialist on my site. It's called the Image Designer, and it does exactly what hero-generator does on my local setup, except you don't need a local setup. You don't need Claude Pro. You don't need a Vercel account. You paste a prompt. You get a sophisticated cover image.
Find it here: the Image Designer specialist.
How it works in plain language: you tell it what you're building (a SaaS product, a portfolio piece, a launch announcement, whatever). You describe the vibe (minimal, brutalist, premium, playful). It generates the HTML/CSS that, when rendered, looks like the cover image you wanted. It screenshots the render. You ship.
If your use case is more specific, I also built two narrower variants. The LinkedIn Image Prompt is tuned for the 1200x627 LinkedIn post format with copy hierarchy that actually reads on a feed scroll. The Instagram Image Prompt is tuned for the 1080x1080 square format with the kind of high-saturation, scroll-stopping treatment that works on a phone screen. Both are siblings of the Image Designer. Same primitive. Different aspect ratios and design constraints.
But if you only click one thing, click the Image Designer. It's the most general-purpose of the three, and it's the closest in spirit to what Claude Design will eventually do for you, if you're willing to pay for the privilege.
Why I Keep Solving Problems Where the Field Is Going
Here's the bigger pattern I want you to sit with.
I didn't set out to predict Anthropic's roadmap. I set out to solve a problem nobody else seemed to be solving for the kind of person I was at the time. Someone with no design budget. Someone who needed sophisticated cover images for a portfolio she was preparing to send to recruiters. Someone who was trying to apply for jobs and didn't want to pay $20 a month for a Canva Pro plan that wouldn't even produce the output she wanted.
I built the workaround because I needed it. Then I documented it because writing it down made the methodology real. Then I packaged it as an AI Specialist because I realized other people had the same problem. And then, five months later, Anthropic shipped a feature that does the same thing for paying customers.
That's not magic. That's what happens when you solve real problems instead of waiting for the perfect tool.
I Build Before the Room Is Ready
When I was 16, I wanted to record math videos for YouTube. I had no quiet room, no soundproofing, no idea noise-cancellation software existed. I had pillows. So I built a fortress out of pillows, shoved my head inside it, and recorded.
I did not wait for the right environment. I did not wait for permission. I built with what was in the room.
That is the only thing that happened with Claude Design too. Anthropic had not shipped the official tool yet. I needed sophisticated images for my portfolio projects. So I used what was in the room: Claude, screenshots of HTML pages, and a workflow I figured out alone. The official feature did not unlock the work. The work was already happening.
If you have been waiting for Claude Design before you start making sophisticated images, you do not need to wait. The primitive has been there the whole time. You can use the workaround today: the Image Designer. And if you want to see the other patterns I shipped before they became official features, they live on my case studies page.
The 16-year-old with the pillows did not know she was a builder. She just built. I am still her.
Want more than cover images? Banners and carousels live on ViteHero.
As always, thanks for reading.