Select a design and suggested next steps appear to click

Non-designers stall on AI design tools for the same reason: a blank prompt box and a blinking cursor, waiting for words they do not have. You know this version is wrong, but you cannot say "hero is too dense" or "hierarchy is flat," and "try again" is the only request you know.

Replit Design flips that. Generate something, select it, and a panel floats up: Suggested next steps—a few color cards, one plain sentence each. Click one and a new design lands beside it; the original does not move. Officially this is Ambient Intelligence, and in the product docs it is the centerpiece of the whole UI.

Suggestions come in three flavors:

What matters is what the click does: the new design lands next to the original, and the original stays—a fork, not a replace. Wrong branch? No need to undo. Two versions sit side by side; you pick with your eyes. Docs call it exploring by direction: following one path is not committing to it.

Cards and typing work together: cards explore wide; typing dials in precision. Docs note that even "move the nav left" opens a new direction—chat is not only for final polish.

Templates are not just for kickoff—you can drop one in mid-canvas

Suggestion cards grow from what you already have. Where does the first frame come from, and what if you stall halfway? Replit Design has four entries: type a brief, pick a template, import something existing, or start from a saved design system. The last two get their own sections; this one is templates.

A library panel sits on the right with three tabs: Templates, Inspiration, and Library (what you have already made in this project). Templates are searchable and filterable—docs list posters, social posts, landing pages, mobile, banners, prototypes; the demo also shows email. Who made them depends on the source: docs say Replit; the launch page says hundreds of real designers. Click or drag onto the canvas and it lands as a live, editable frame—same as anything you generated, ready to edit or Build into an App.

On the launch page, Replit draws a line against peers: other AI design tools treat templates as a starting gate you either use or skip at the beginning. Here they are ingredients—you can drop one in mid-flow. The demo uses them that way.

Built-in Mobbin screens: great for reference, never your code

The same panel has a second tab that looks like Templates and behaves completely differently. This is the easy place to misread the product.

Inspiration is powered by Mobbin: a library of real App UI screenshots, built since 2018 per its own site. Replit calls it the world's largest UI/UX reference library—2 million designers, weekly updates, 1,000+ Apps, 600k+ real screens—and claims a first: no Mobbin account required; it is already wired in.

For scale: Mobbin still sells its own subscriptions today—about $10/month annually for individuals, $16/user/month for teams. Free is tight: only the newest 4 Apps and 4 sites, and at most 3 collections. So searching without an account is a real perk if you never paid.

This is easy to misread. The launch page says you "draw from 600k+ real screens to turn ideas into finished Apps or sites," which sounds like screenshots become your page. Product docs shut that door: inspiration images add no code to your project. They are pictures for you and the model to look at.

How Mobbin and Replit relate is not on the launch page—acquisition, investment, or licensed partnership is never stated, and public reporting has no acquisition notice. What is clear: Mobbin's own site still sells subscriptions. Library size also disagrees: Replit says 1,000+ Apps and 600k+ screens; Mobbin's pricing page meta says "hundreds of apps," and its about page still lists 400k+ users and 100k+ screenshots. This piece uses Replit's figures.

Brand consistency sits in a style file both people and models can read

References and templates answer whether one frame looks good. By frame five or ten, the question is whether they look like one brand. Replit's launch diagnosis: today's AI design is fragmented—prompt in one tool, edit in another, ship in a third—and each handoff loses something. Colors drift, fonts reset, and you ship a diluted version.

The fix is a built-in design system. Strip the marketing and you get five things inside a generated system:

Storage is concrete, not a hollow brand PDF: tokens live in tokens.json, feed CSS variables (e.g. src/index.css), and components consume those variables. It also writes a DESIGN.md—the same system as a document both you and the AI can read.

Two ways to build one. From scratch: brief it on the Replit home page and answer a short brand questionnaire. Or extract backward from a design you already like—three-dot menu on the frame, extract, and it pulls color, type, and spacing into a stored system. Seeds can also be a Figma link (the interface design tool designers often use), a live site, a code snippet, reference images, or an existing DESIGN.md. Saved systems live in your workspace; you can pick one for new work and swap per project.

One click builds an App from a design—free tier stops there

If a design's destination is something that runs, what does that last step require?

Select a design frame and its action bar has a Build menu with two paths:

The first creates a real App from the mock—layout, styles, assets become a running program in the project's work list. The second lists every App already in the project; pick one and the AI rewrites that App's code to match the design, leaving the mock on the canvas as reference. Reskinning finished work uses this path. Both auto-save a checkpoint before they run so you can roll back.

Here is the catch. On the launch page, access is one line: open to everyone today. The official billing table looks like this:

Docs say: building a new App from a design needs Core or Pro; applying a design and every other Design capability work on all plans. So of the two paths above, only the first is gated—applying a design onto an App already in the project works on free. Free can generate, fork, click-edit, and export PNGs within the daily allowance; it just cannot open a brand-new running thing from a mock.

Two notes on how credits burn. AI is metered by effort: a small bug is cheap; a whole feature is expensive; and any chat costs money, even if the reply is pure text and zero code changes. The saving grace: simple edits in the click-to-edit panel (copy, color, spacing) write source directly and cost no credits—only changes with hidden complexity get handed to the AI. One more limit: that panel is browser-only; mobile and desktop Apps do not get it.

You can swap models anytime—five on the launch page, four in the docs

If models do the designing, swapping a model is swapping a designer. Official docs treat that as its own exploration path: suggestion cards are the same designer trying a new angle; a model swap is a different person. Same prompt, two models, two versions side by side.

The two sources disagree. Both, as published:

The docs list is the main model set under Power mode in the model-picker screenshot. It also says the roster changes as new models ship—trust what your account's in-product menu shows.

Asset generation uses a separate model stack from the design model:

Nano Banana is Google's image model; vectors also list Gemini, so the launch page's "Gemini" likely lives here. GLM has no matching entry on either list.

Two other knobs in the same dialog: mode—how you trade speed, cost, and capability, listed here as Lite, Economy, and Power (billing docs also mention Turbo, Pro-only); and how hard it should think, Low to Max, each model with its own recommended default.

Figma frames, Claude designs, and URLs all land as starting points

Everything above assumes you start from zero. If you already have something? Four channels:

URL import clones a live site into a design mock—your current homepage, or a page you want to redo. Figma import rebuilds a frame as a design. Both arrive with suggestion cards. Doc examples differ slightly: Figma gets "dark editorial variant," "add a pricing section," "mobile version"; URL gets "dark editorial variant," "mobile version," "add a contact page."

The Claude path runs the other way

Importing a design from Claude is not started in Replit—it is initiated in Claude: Share → Send to… → Send to Replit. After you authorize, Replit creates a project, drops your design as a static index.html, then auto-starts a task called "Turn this Claude design into a real app" to make that static page interactive. Docs note this is a new integration; whether you see the option in Claude is controlled on Claude's side, and not every account has it.

Two small utilities: for other screen sizes you do not hunt settings—ask the AI for a mobile view and it renders at 390×844 (tablet 768×1024, desktop 1280×720). Design and App frames can one-click save as PNG (image/video media frames download the source file); a shortcut copies the selected frame straight to the clipboard as an image.

Early users shipped posters, invitations, and store pages

Mechanics covered, a last look at real output. The launch page opens with a hero video; the first screen recording after the body is a waterfall of early-access work.

Replit's claim: everything from marketing assets to live-shipped Apps and sites. Two early users:

I run a brand strategy firm, and I have very clear taste. Most AI tools spit out something too generic to put my name on. Replit Design genuinely surprised me—I'm impressed, and I'm only getting started.

Darby IRL, brand strategy founder

I can stay in the creative work instead of getting stuck on development. Projects I tried before and never shipped can go live now.

Agatha, Designely

Open to all users today at replit.com/design. They are also running Designathon—a design marathon with a prize pool over $50,000 in cash plus Replit credits for work that "makes them stop in their tracks." Last month they hosted an event in New York called vibecon.

People who don’t design, stuck in AI design tools, usually freeze at the blank prompt: you know this version is off, but you can’t name “hero is too dense” or “not enough hierarchy.” Replit Design flips that. Generate something rough first, select it, and a strip of cards floats up—each one a plain-English nudge. Click one and a new mockup grows beside the original; the original stays put. Replit calls this suggestion system Ambient Intelligence.

There are four ways into Replit Design: type a description, pick a template, import something you already have, or start from a saved design system. The library panel on the right of the canvas puts two tabs side by side. Templates are Replit’s own starter frames; Inspiration is backed by Mobbin, a reference library of real App UI screenshots—Replit says 1,000+ Apps and 600,000+ screens. Mobbin still sells its own subscription today: personal annual works out to about $10/month, and free only shows the latest four Apps.

Templates and references handle whether a single screen looks good. By the fifth or tenth frame, the question becomes whether they look like they came from the same brand. Replit Design stores the brand as a file both humans and machines can read: change the bottom layer, and everything above follows.

Select a mockup and the Build menu in the action bar offers two paths: spin up a new runnable App from it, or apply it onto an App already in the project (the mockup stays on the canvas as reference). Both auto-save a checkpoint first so you can roll back. The launch page only says it’s “open to all users today.” Replit’s billing docs spell out the table: canvas, click-to-edit elements, and AI chat work on every plan; spinning up a new App from a design needs Core or Pro—of the two paths, only the first is gated.

Sources for the numbers on this page: library scale (1,000+ Apps, 600,000+ real screens, 2 million designers) is Replit’s claim; Mobbin’s own site currently lists hundreds of Apps. Plan gates, prices, and billing rules come from Replit product docs and the billing page—not the launch page. Demo timings and amounts are from Replit’s official recordings. Model lists also disagree: the launch page names Claude, GPT-5, Gemini, Kimi, and GLM; current docs list Claude Fable 5 (recommended), Claude Opus 5, GPT-5.6 Sol, and Kimi K3, and note the list changes as new models ship. Entry point: replit.com/design.