Anthropic launched Claude Design on April 17, powered by Opus 4.7, turning prompts into prototypes, slides, mockups, and decks
Design system integration reads your codebase and design files, extracting tokens like colors, fonts, spacing, not logic
Four export formats ship: PDF, URL, PPTX, or push to Canva fully editable with the Canva Design Engine underneath
Research preview is live for Claude Pro, Max, Team, Enterprise, no separate price, consumes your plan's usage limits
My take: not a Figma killer, it is a draft generator you hand off to Canva or Figma to finish
Anthropic shipped Claude Design yesterday. Not a waitlist. Not a research paper. A real product that turns text prompts into prototypes, slides, pitch decks, and landing page mockups, then lets me refine them through conversation, inline comments, direct edits, and custom adjustment sliders Claude generates on the fly.
I have been poking at it for a few hours. Here is everything Claude Design does, what is happening under the hood, and who actually gets access today.
What Claude Design Actually Does
The pitch is simple. I describe what I want. Claude builds it. I refine it.
The example from the launch demo was "prototype a serene mobile meditation app, calming typography, subtle nature-inspired colors, clean layout." Claude returned a full mobile app mockup in seconds. Multiple screens with typography that was not random and a color palette that felt deliberate. From there I asked for a dark mode toggle, a different accent color, tighter spacing. Claude adjusted without starting over.
The revision model is the interesting part. I get four ways to refine a draft.
Conversation works like any Claude chat. Type what to change. "Make the headline bigger, swap the accent to warm orange, drop the testimonials row."
Inline comments let me drop a note on a specific element, the same way I would review a Google Doc. Click the button, write "this padding is too tight," Claude fixes that one element without touching the rest.
Direct edits let me change a value, move a block, or replace text without waiting for a round trip to the model. Instant.
Adjustment sliders are the part that feels new. Claude generates custom sliders for whatever I am editing in the moment. Font weights, hue shifts, corner radius, layout density. The slider only exists because I asked about that specific property. Most AI design tools give me a textbox. Claude Design gives me knobs that only appear when relevant, then disappear when the conversation moves on.
All of this runs on Opus 4.7, which shipped earlier this week with 3x the image resolution of previous Claude models. That matters here. Claude can see its own output in high fidelity, which means revision quality holds up as I iterate deeper into a design instead of collapsing into mush by iteration five.
How It Uses Your Design System
If you are a solo maker, the default output is fine. If you are a team with a brand, this is where Claude Design earns its keep.
I can hand Claude my codebase and my design files during onboarding. It reads them. It extracts design tokens: colors, fonts, spacing values, visual components. It does not read my application logic, and code beyond the token extraction is not exposed. The tokens become a system Claude applies to every future project I generate in that workspace.
That means I can ask for "a landing page for my new onboarding flow" and get a draft that already matches my brand. Colors correct. Type scale correct. Button shapes correct. Spacing grid correct. No fiddling to put the accent color back to my lime green after Claude picks purple on its own.
Onboarding is live and interactive. Claude pings me with a quick question when the extracted tokens look ambiguous. Which of these three grays is your background versus your surface color? Is this border radius intentional, or a coincidence across three components? Answering those questions once tightens every future output. I probably spent 15 minutes total on the initial setup and saved hours of correction on every draft after.
Teams can maintain multiple design systems, which matters if you work across brands or run an agency. You can refine individual components, override specific tokens, and keep governance rules (minimum font sizes, accessible contrast ratios) consistent across everything Claude builds.
This is the enterprise hook. Figma gave teams shared libraries. Claude Design gives teams shared libraries that auto-apply to AI-generated work. Different layer of the stack, same problem solved differently.
Export Options and the Canva Angle
The output ships in four formats, and this is where Anthropic got clever.
PDF covers email, Notion embeds, and anything that needs to render without software. URL gives me a hosted link to the live prototype, ideal for async feedback from stakeholders who hate installing things. PPTX is PowerPoint, because corporate workflows refuse to die. And "send to Canva" is the format that changes the strategic picture.
Claude Design is built on Canva's Design Engine and Visual Suite. When I send a file to Canva, it opens as a fully editable, collaborative design. All layers stay intact, text stays editable, and the full Canva asset library is available to extend the output. That library runs about 200 million items across stock photos, icons, fonts, and video, so the extension surface is huge.
The pattern Anthropic is betting on looks like this. Claude generates the draft. I polish in Canva. Neither product tries to own both sides of the workflow. That is why Canva's VP of AI showed up in the launch materials calling the partnership complementary rather than competitive. The deal has real mechanics behind it, not just a logo on a slide.
If you already pay for Canva Pro, this is basically a free upgrade. You get an AI drafting layer in front of tools you already know, with no extra subscription.
Who Gets Access and What It Costs
Here is the part everyone asks first.
Claude Design is rolling out as a research preview. Some features still feel rough, and Anthropic may change how they work based on what breaks in production.
Paid Claude plans get access. Pro for individuals at 20€ per month, Max for heavier individuals, Team for small orgs, Enterprise for the big checks. Free tier users do not see it. API-only customers do not see it either. This is a consumer product, not a developer endpoint.
Price is zero on top of your plan. Claude Design consumes your existing usage limits. On Pro, Claude Design shares the same message budget as regular chat. A complex design request chews through your allowance faster than a casual conversation, which is worth knowing before you burn a day prototyping.
For Team and Enterprise plans there is no per-seat upcharge. Design system onboarding, where Claude reads your codebase and files, is a one-time setup per system. Generating designs against it afterwards is normal usage.
Rollout is gradual. If you are on a paid plan and do not see Claude Design yet, Anthropic is trickling access throughout the day. Refresh claude.ai, check the sidebar, wait a few hours. It will show up.
One more detail worth flagging. The research preview label is not cosmetic. I hit a few rough edges during my testing sessions, mostly around very long prompts or designs with more than 20 elements. Claude occasionally lost track of specific tokens halfway through a revision. It recovered fine when I pointed the mistake out, but expect to proofread more carefully on longer projects until the preview flag comes off.
Bottom Line
Claude Design is not a Figma killer. I do not think Anthropic wants it to be.
What it is: a drafting tool that skips the blank canvas problem. Founders without a designer get pitch decks that do not look like homework. PMs waiting on the design team get clickable prototypes to share with stakeholders the next morning. Solo marketers can ship campaign one-pagers without ever opening Figma.
Designers still win on the finishing work. Claude spits out a draft in 30 seconds. Taking that draft from "fine" to "shipped" still requires taste and edge case judgment that humans do better.
The Canva integration is the underrated piece. It means Claude Design's output lands in a tool 170 million people already use, with collaborators they already invited. Zero context switch. That alone is why I think this product sticks while a dozen AI design tools before it did not.
Worth opening Claude tonight if you are on Pro or Max. Start with a pitch deck. See how close the first pass gets. Adjust your expectations based on real output, not launch day hype.
Top comments (0)