Google Stitch

✨ Overview

Google Stitch is a free AI UI design tool from Google Labs, at stitch.withgoogle.com. It started life as Galileo AI, which Google acquired in 2025, and relaunched under its current name at Google I/O in May 2025. The pitch fits in one sentence: describe an interface or an app in plain English, and it hands back a high-fidelity UI prototype plus frontend code. You can also feed it an image, a sketch, or your voice. No design background required.

The name is the thesis. Stitch, as in sewing. You bring the idea, it pieces the interface together. Google calls the approach 'vibe design' — you describe the business goal, the feel, or a visual reference, and the model does the layout work that used to start with wireframes.

Start from 'what should this feel like', not 'where does the button go'.

For US teams, that's the hook: it shortens the gap between a vague idea and something you can click, show a stakeholder, or hand to a designer.

🧩 Key features

Stitch has had two distinct eras. At launch in 2025 it was a single-screen generator. The March 2026 update turned it into an AI-native design canvas. The base features first:

• Text-to-UI: one sentence, one interface.
• Image, voice, and sketch input: drop a screenshot or a napkin sketch as a visual anchor.
• Natural-language iteration: 'make it dark mode', 'add a modal' — no need to re-describe the whole screen.
• Figma export: with Auto Layout, named layers, and editable components.
• HTML/CSS export: Tailwind CSS and Google Fonts, ready for a frontend.
• Public sharing: added October 16, 2025, one-click share links.

The March 2026 release raised the ceiling:

• Multi-screen generation: up to five linked screens in one shot.
• Infinite canvas: spread every screen out and arrange them.
• Voice Canvas: edit and critique designs by talking.
• Interactive prototypes: connect screens, define transitions, hit Play to preview the flow.
• Design-system import: pull in a spec via URL or a DESIGN.md file.
• Agent Manager: explore several directions in parallel.
• Export expansion: Figma, HTML, Tailwind, React, ZIP, MCP, AI Studio, and Jules.

🤖 AI & the models underneath

Stitch runs on Google's own Gemini family. Early versions used Gemini 2.5 Pro and Flash; later builds moved to Gemini 3 Flash and 3.1, with Nano Banana handling the design repainting — redrawing the first pass into something cleaner. You don't pick a model; it happens behind the scenes, and the results just get better over time.

What you do pick is the mode, and each mode maps to a different model strategy:

• Standard: faster, exports to Figma, fine for most tasks.
• Experimental: runs Gemini 2.5 Pro, higher fidelity, accepts image input, exports HTML/CSS but not Figma.

Everything flows from 'vibe design'. Stitch wants to be an idea amplifier: you supply the direction and the feeling, it supplies the pixels. That positioning is why it feels like a fast output tool rather than a fine-grained pixel editor.

🛠️ How to use it

Nothing to install. Open stitch.withgoogle.com, sign in with a Google account, and you're in. The usual path:

• Pick a mode. Not sure? Start with Standard.
• Describe the screen. Lead with the business goal — 'a booking flow for a yoga studio, calm, lots of whitespace' — rather than pixel specs.
• Add a reference if you have one: a screenshot, a sketch, or just talk at it.
• Read the result, then iterate in plain language: 'swap the primary color for deep green', 'add a fourth tab to the bottom nav'.
• Export. Figma for refinement, HTML/CSS to hand off to a frontend.

One caveat for anyone outside the US: access depends on a Google account, and the tool assumes an English prompt for the best results.

💡 Tips & best practices

• Describe the goal, not the pixels: vibe design responds to intent. 'A checkout page that makes ordering feel effortless' beats '40px margins on the left'.
• One change per iteration: edit a single thing at a time. Five stacked requests muddy the output.
• Anchor with a reference image: a screenshot of an app you like pulls the style much closer than words alone.
• Lock the design system with DESIGN.md: if your team has tokens for color and type, import them once instead of restating every time.
• Treat Stitch as the start, not the finish: export to Figma and keep going. Auto Layout and named layers are already there.
• Explore in parallel: Agent Manager can run several directions at once; keep the best, discard the rest.
• Prototype before you commit: multi-screen generation plus interactive previews let you click through a flow before a single line of production code.

🌍 Adoption

Google doesn't publish Stitch's signup or monthly-active numbers, so treat any specific figure as fiction. What's verifiable is momentum: the day of the March 2026 update, Figma's stock dropped roughly 8% to 9%. A free tool shipping a major release was enough to move a multi-billion-dollar design company — the market read it as a real threat.

In the US ecosystem, Stitch sits at the top of a funnel that ends in Figma and production code. Designers use it for a first pass, PMs use it to validate an idea before a sprint, and founders use it to show investors something clickable instead of a slide. The community has largely settled on 'Stitch to explore, Figma to finish'.

💰 Pricing

Stitch is free. No paid tier, no trial countdown. The free allowance is roughly 350 standard generations plus 200 experimental generations per month. That's plenty for personal prototyping and early ideation; if you're producing at high volume you'll hit the cap, and there's no paid expansion available yet, so pace yourself or wait for Google to formalize pricing.

Free doesn't mean zero cost, though. What Stitch hands you still needs accessibility work, edge-case handling, backend integration, and hardcoded values swapped out. Budget for the human hours, not just the tool — the tool is the cheap part.

⚖️ Pros, cons & limitations

Pros

• Zero barrier: high-fidelity UI without design skills.
• Multimodal input: text, image, voice, sketch all work.
• Useful exports: Figma with real layer structure, HTML/CSS on Tailwind.
• Fast iteration: 'make it dark mode' is a valid edit.

Cons

• It generates UI, not a working app. Early builds were static states with no real interaction; even today's interactive prototypes are a long way from shippable product.
• Not a Figma replacement: no mature plugin ecosystem, no co-editing, no offline mode.
• You still do the unglamorous work: accessibility, edge cases, backend integration, replacing hardcoded values.
• Highly original designs and big projects (10+ screens) strain it.
• Best results come from English prompts.

🔄 Stitch vs. the alternatives

Against Figma, the relationship is clean: Stitch is the on-ramp, not the rival. It's built to hand off — the Figma export carries Auto Layout and named layers precisely so you keep working there.

Against code generators like v0, Lovable, Replit, or Cursor: those optimize for 'code that runs'. Stitch optimizes for 'what the thing looks like'. It exports HTML/CSS/React, but its first job is the design, not the runtime.

Against other AI design tools — Uizard, Framer AI, Recraft — Stitch's edge is the Gemini backbone and the price: a genuinely free tier with no friction. Galileo AI, its own predecessor, is now folded into the product.

The honest summary: Stitch wins the 'first five minutes' of a project. It doesn't try to win the last mile.

❓ FAQ

What's the relationship between Stitch and Gemini?

Stitch is the product; Gemini is the engine. Under the hood you're getting Gemini 3 Flash, 3.1, Nano Banana, or 2.5 Pro depending on the mode.

Is it really free?

Yes. Roughly 350 standard plus 200 experimental generations a month, no card required.

Can it export to Figma?

Standard mode can, with Auto Layout, named layers, and editable components. Experimental mode skips Figma and exports HTML/CSS instead.

Does it build a complete, usable app?

No. It produces UI and prototypes, not production software. Backend, accessibility, and edge cases are on you.

Will it replace Figma?

No. It's the entry tool that feeds work into Figma, not a substitute for it.

Does it work in languages other than English?

It's improving, but English prompts still give the most reliable results.