Google adds Stitch, a browser AI canvas to speed starter UI design: what changed

Stitch is a free beta browser — based AI design canvas from Google Labs that generates multi — screen starter UIs from natural — language prompts, exports to Figma or other formats, and manages a project — level DESIGN.md.

Google has released Stitch in beta through Google Labs, a browser — based AI design canvas intended to speed the creation of polished starter screens for product mockups. The tool is aimed at builders who want usable initial UIs quickly without building wireframes by hand, mastering Figma, or hiring a dedicated designer on day one.

Stitch traces its lineage to Galileo AI, which Google acquired in 2025 and reintroduced at Google I/O that year. The system runs on the Gemini model family and applies smaller remixing models such as Nano Banana to adjust and iterate screens. For each project Stitch also generates a project — level DESIGN.md file to serve as a single source of truth and offers export paths that let developers push generated assets directly into Figma or download other formats from the browser.

The tool’s user flow is text-driven: users supply natural — language instructions, Stitch builds a design system from those prompts, and then renders multiple starter screens on a canvas for side-by-side comparison. Generation typically takes a few minutes. Individual screens can be opened for in-place editing, annotated, or refined with local prompt windows, and a Live Mode (currently in preview) accepts chat and spoken instructions that queue changes for the AI to execute.

Stitch is free during the beta and operates on a daily credit allotment reported at 400 credits per day. In one test, creating a new project — a base design system plus five screens — consumed roughly 9 credits, while individual edits fell between about 2 and 5 credits each. Those per-action costs and the daily cap will be important for teams planning repeated iterative loops or high-volume prototyping.

Technically and commercially, Stitch marks a step from earlier single — screen AI outputs toward multi — screen, design — system-aware generation that appeared in 2026. The models reason about layout and component choices using common design patterns, which speeds creation and yields functional starter UIs but can bias results toward conventional templates. Producing unique visual systems may therefore still require careful prompt engineering or manual design work.

For practical workflows, Stitch’s DESIGN.md provides consistent context for edits and new screens much like AGENTS.md does for coding agents, and Figma integration lets teams fold generated assets into existing processes. Teams should plan for current limitations: Live Mode remains in preview, the beta credit cap may constrain heavy prototyping, and model — driven defaults often need refinement to achieve novel or brand — specific visual results.