Stitch AI Design Tool: Agentic System, Voice Design, Free Export

Google Labs has released a major update to Stitch, its generative AI design tool. The platform has moved from a simple screenshot utility to a full‑on competitor to AI‑driven design solutions such as Figma. Central to the upgrade is an agentic system that incorporates recent Gemini text and image models, allowing the tool to interpret prompts and construct designs autonomously.

New Features and Capabilities

A native design canvas now lets users prompt the system to build complete layouts, much like AI Studio generates code. The “design agent” can be instantiated multiple times and runs on either Gemini 3 flash or Pro models, giving flexibility in speed and determinism.

A new design.md file acts as a portable design system toolkit; it can be edited graphically or directly in code editors and applied across multiple projects. By providing a website URL, Stitch extracts primary and secondary colors, fonts, and styling cues such as icon and button designs, then incorporates those standards into the generated design.

Export options have expanded to include direct delivery to AI Studio for code generation (including authentication and database scaffolding), Figma, and React applications. The tool also supports instant mockup prototypes that wire navigation between pages automatically.

Voice interaction is enabled through a “vibe design” feature that uses Gemini Live models, allowing users to make design changes with spoken commands. All of these capabilities are offered for free, including the tokens required for model usage.

Demonstration and Examples

In a typical workflow, users select a mobile or web app canvas and either type a prompt or paste a URL. A demonstration used the URL of a Thai resort site to capture its visual vibe. Running on the Gemini 3 flash model, Stitch identified the site’s brown tones as primary colors, extracted complementary secondary colors, and recognized the fonts used for headings and buttons.

The system then generated initial designs, showing a side‑by‑side comparison of outputs from the flash model and the Pro model. While the Pro model tended to be more deterministic, the flash model produced strong results in this case. Users could request specific pages—such as programs, cuisine, or residences—and preview them on desktop, iPad, or phone layouts.

An instant prototype was created with navigation wired between pages, and users could modify elements on the fly using AI or connect them to existing screens. By asking for a “more holistic natural food look,” Stitch produced three variations of a cuisine page, inserting placeholder images generated by the Nano Banana 2 model.

Advanced Features and Export Workflows

The “live mode” leverages Gemini live bidirectional models for voice‑driven design, enabling real‑time adjustments through spoken commands. When exporting to AI Studio, users can supply prompts such as “make this real and add a user dashboard with dynamic data,” prompting the generation of functional React.js code, HTML, and image assets.

Export pathways also include Figma and direct React app generation, as well as instant mockup prototypes. A project brief can be exported, producing a product requirements document that contains the design system and color palette. The design.md file can be accessed, edited, and imported from other websites, allowing designers to reuse or adapt existing brand standards across new projects.

Overall Impact and Future Potential

Stitch now positions itself as a strong alternative to Figma, especially for users who are not professional designers but have clear visual references. Google Labs’ close integration with Gemini models from early development stages enables rapid prototyping and product iteration. The speaker envisions broader applications such as YouTube thumbnails and general graphic design, suggesting that future extensions may go beyond apps and websites. All of these features are available at no cost, encouraging widespread experimentation.