Google Stitch · 2026 Hands-On Review

Google's Vibe Design (Stitch 2.0): Tested Hands-On

Dhairya Purohit

Builds dMaya. Ships AI design workflows in real client work.

Published April 29, 2026

Google Labs introduced "vibe design" with the Stitch 2.0 launch on March 18, 2026. The post was titled Introducing 'vibe design' with Stitch. The framing: instead of starting with a wireframe, you describe a business objective, a feeling you want users to have, or an inspiration. Stitch generates high-fidelity directions on a canvas. Voice input. Free during Labs. Six weeks later, every AI design tool in the category claims vibe design.

We tested Stitch the same week. Same prompt across Stitch (Flash and Pro), Claude Design (Opus 4.7), and dMaya (Opus 4.7). Same evaluator, same hour, same brief. This is the Stitch-specific write-up: what it does well, where it breaks, the real meaning of 350 generations a month, and which tool to reach for when Stitch is not enough.

We are dMaya, one of the tools in this category. The comparisons below are honest about Stitch, including the things it beats us on (speed, voice input, free during Labs).

What Google means by "vibe design"

From the official Stitch 2.0 launch post: "Instead of starting with a wireframe, you can start by explaining the business objective you're hoping to achieve, what you want your users to feel, or even examples of what's inspiring you." That sentence is the closest thing the category has to a canonical definition.

Three things in that framing matter and are unique to Stitch's pitch. First, prompt-first instead of layout-first. Second, intent (objective, feeling, inspiration) instead of components. Third, exploration of multiple directions in parallel rather than committing to one wireframe and iterating it.

The category itself now includes Claude Design, dMaya, Figma Make, and Lovart. Google coined the term, but the practice exists across all of them. For the broader picture, see our vibe design field guide. This page is specifically about what Stitch does.

Stitch 2.0's six vibe design features

The March 2026 update added six things that are genuinely new to the product, separate from the broader Stitch interface that existed before.

Two of these (Voice Canvas and design.md) are genuinely innovative in the category. The other four are table stakes that other vibe design tools also offer.

Hands-on test: same prompt, real output

We ran Stitch on a freelancer SaaS dashboard prompt on April 24, 2026, alongside Claude Design and dMaya on the same brief. Below is the Stitch (Pro) run. Full three-tool test with all videos and timings lives at /blog/google-stitch-vs-claude-design-vs-dmaya.

Stitch produced output in two minutes on both Flash and Pro. The output had recurring positioning issues (especially with center-aligned elements), and screen-to-screen consistency was absent because each generation is independent. The output was useful as early exploration. It was not a deliverable.

Where Stitch wins

Three things Stitch genuinely does better than every alternative.

• Speed of first output. Two minutes from prompt to a visual you can react to. Faster than Claude Design (~10 min) and similar to dMaya (~2.5 min). For early ideation when you just need something on screen, this matters.
• Voice Canvas via Gemini Live. Genuinely novel. The voice loop for cosmetic adjustments works and feels different from typing. Best implementation of voice editing in the category right now.
• Free during Labs. No financial friction for testing. The 350/month cap is real, but for hobby work or evaluation, the price is right.

If your work is genuinely exploratory and the deliverable is "a visual to discuss" rather than "a design to ship", Stitch is the right call.

Where Stitch breaks (5 concrete limits)

1. Output positioning issues

The most consistent failure mode is alignment. Center-positioned elements drift. Hero headlines miss their grid. Asking Stitch to fix the positioning often produces a different layout entirely rather than a corrected version of the previous one. LogRocket's published Stitch tutorial flags the same issue independently.

2. Visual sameness across generations

Run the same kind of brief twice and the outputs share structure: similar hero patterns, similar button styles, similar typography pairings. Stitch converges on the output distribution Gemini was trained on. The advertised "multiple directions in parallel" produces variations within a narrow band, not genuinely distinct visual languages.

3. No design system or token integration

Stitch's design standard extraction handles surface-level brand consistency (color and typography) by analyzing URLs or assets. It does not load existing design tokens, component libraries, or motion rules. For agency teams with a real brand system, every Stitch generation invents a fresh visual identity that needs to be hand-mapped to the existing one.

4. Single-user only

No multi-stakeholder collaboration in the canvas. You can share an export, but you cannot co-edit, leave inline comments, or run a real review session inside the tool. For teams used to Figma's collaborative cursor model, this is a regression.

5. HTML/CSS export only (no clean framework conversion)

Stitch exports rough HTML and Figma frames. There is no first-class React, Vue, or framework export. To get from Stitch to production code, you hand the HTML to a coding agent (Claude Code, Cursor, Codex) and ask for the framework conversion. Possible, but it is a gap the user closes, not a path Stitch ships.

350 generations a month: real workflow math

Stitch is free during Labs, with caps. The Standard tier allows 350 generations per month. Experimental allows 50. The headline number sounds large until you do the workflow math.

In our hands-on testing and in published Stitch tutorials, one usable design typically takes 3 to 5 generations of iteration: an initial pass, one or two adjustments for positioning and structure, often a re-roll for a different aesthetic direction. Three screens with that depth eat 9 to 15 generations.

At that rate, 350 generations is roughly 70 to 100 finished designs per month. For a solo builder, that is plenty. For an agency running multiple client projects in parallel, it is tight. For anyone doing real iteration on a complex product, it can disappear in a single sprint.

The cap also lacks rollover. Light months do not bank generations for heavy months. Once you hit zero, you wait for the calendar.

What to use when Stitch is not enough

Three alternatives matter, depending on what Stitch is failing to do for you.

Lovart and Visily are also in the broader vibe design conversation, but they target marketing and wireframes respectively, not product UI. See the vibe design field guide for the full landscape.

When Stitch is the right call

Stitch is a real product doing real work in the category Google named. It is also deliberately positioned at exploration, not delivery, and the workflow gap between Stitch output and shippable product UI is wider than the marketing implies. For solo work and ideation, it is the right call. For everything else, you are reaching for a different tool.

We built dMaya because the workflow Stitch points at, prompt-first design with a real canvas, deserves to ship product UI on a budget that does not break at scale. If that is the work you are doing, the link below is the one we want you to follow.

Keep reading

The full test

Google Stitch vs Claude Design vs dMaya

Same prompt, three tools, real timings on video.

The category

Vibe Design: The Definitive 2026 Guide

Definition. 5 tools compared. Real timings. The workflow that ships.

Alternative

The Best Google Stitch Alternative in 2026

If Stitch breaks on multi-screen or export, here is what to use.