Field notes: Google Stitch, the free Gemini UI generator
Google Stitch is a free Google Labs experiment that turns a text prompt — or a photo of a whiteboard sketch — into a UI design plus working frontend code, powered by Gemini. I build Drupal themes for a living but nobody would mistake me for a designer, so a machine that produces plausible-looking interfaces on demand is aimed squarely at people like me. I gave it a proper look.
What it actually is
Stitch was announced on the Google Developers Blog around Google I/O in May 2025. It grew out of Galileo AI, a startup Google acquired and rebranded — Galileo reportedly cost $39/month; Stitch is free. You describe an app in plain English, including things like colour palette and mood, and it generates mobile or web screens. Or you upload a wireframe, a screenshot, or a genuinely bad marker sketch and it digitizes the idea.
- Inputs: text prompts, images (wireframes, screenshots, sketches), plus chat-based refinement and theme selectors.
- Outputs: UI screens, a "Paste to Figma" handoff, and frontend code (HTML/CSS at minimum; some coverage reports framework exports too).
- Modes: Standard runs on Gemini Flash — fast, and it's the one with Figma export. Experimental runs on Gemini Pro — deeper reasoning, accepts image input, but appears to hand back only HTML/CSS.
- Cost: free while in Labs, with monthly generation caps — early reviews cited 350 standard plus 50 experimental generations, and later reports suggest the limits have been raised.
The workflow in practice
You essentially write a brief, not a spec. Something like this got me a surprisingly coherent set of screens:
A facility-management portal dashboard for building managers.
Left sidebar navigation, cards for open work orders, a calendar
of inspections, muted blues, dense but readable. Desktop web.Then you iterate in chat: "make the sidebar collapsible", "warmer accent colour". It's fast, and the multi-variant generation is genuinely useful for exploring directions before committing. Since I/O, Google has also been pushing "vibe design": a Stitch Agent that streams its work onto the canvas in real time so you can steer mid-generation, plus sharing via AI Studio and publishing through Netlify.
Against v0, Figma Make and Pencil
Rough positioning as I see it: v0 is component-first — it produces clean React/shadcn code you can drop into a Next.js codebase, and no one else matches that code quality, but it's credit-based. Figma Make starts from your existing Figma designs and respects your design system, but needs a paid Figma plan. Stitch is design-first and free, best when you're starting from nothing — but it knows nothing about your design system unless you describe it in the prompt.
And Pencil, which I've been using lately, is a different animal: it's a local, agent-driven design editor, so my coding assistant designs on an editable canvas next to my actual code instead of in a hosted sandbox. Stitch is better for a zero-to-something first draft; Pencil fits better once the design has to live inside a real project.
The rough edges
The community verdict is remarkably consistent across Hacker News and elsewhere, and it matches mine.
Stitch is a great starting point and a poor finisher — everyone loves the first five minutes and nobody ships the output as-is.
Concretely: the output tends toward the generic — Stitch's designs look like Stitch, and one HN commenter noted it explains "why most vibe coded apps look generic". Reviewers found frequent accessibility misses (colour contrast, touch targets), layouts that are static rather than responsive, and no way to fine-tune a detail without re-prompting the whole screen. As a PHP developer, I also note the obvious: HTML/CSS export is a starting point, not a Twig template.
Still — for free, as a way to get a non-designer past the blank-canvas stage before rebuilding properly in the theme layer, it earns a bookmark. Just treat everything it makes as a sketch, because that's what it is.