Vesper is a night-sky companion that does one thing: tells you what's worth stepping outside for tonight, when to go, and exactly where to look.

Built fully interactively in Google Stitch across 18 screens. A live celestial dashboard, a clockwork orrery with all 8 planets orbiting in real time, constellation deep-dives, ISS pass tracking, moon phase morphing, and a personal observation log. Everything in one place, nothing you don't need.

The aesthetic comes from antique star atlases and brass instruments. Deep midnight canvas, warm gold linework, three typefaces locked to strict roles. No neon. Something that feels like a precision tool you'd genuinely want to return to every clear evening. The Problem:

Most people have looked up at the sky and felt something, then had no idea what they were actually looking at. Existing apps don't help. They're either buried in data, built for astronomers, or so technically overwhelming that curiosity dies before it starts. Nobody has built the calm, beautiful version. The one that just says: here's what's worth going outside for tonight.

The Solution:

Vesper is that app. It starts with the one question that actually matters: what should I look for tonight, and answers it clearly, beautifully, without making you feel like you need a degree first. The design is warm and considered rather than cold and clinical, because a tool you enjoy looking at is a tool you actually use. For those who want to go deeper, the depth is there. But it never gets in the way. How Stitch Was Used:

Streaming generation

Every screen started as a detailed natural-language prompt, watched streaming live onto the canvas. There is something genuinely different about seeing a complex layout appear in real time rather than staring at a blank artboard. The Tonight dashboard, the Orrery, the sky gradient on Sun and Twilight. All of them generated in one shot from a single prompt. That moment is worth capturing in any walkthrough video.

In-place AI edits

After the initial generation, individual elements were refined by clicking directly on them and describing the change in plain language. Things like "make the meteor streaks fire in groups of three with different angles", "add a real dark-mode map with a street grid", "set Sky Map as the active sidebar item", and "replace the static stars with the same drifting animation as the Welcome screen". This is where Stitch earns its keep. It stops feeling like a design tool and starts feeling like a conversation. You say what you want, you watch it happen, you iterate.

Motion and shader pass

Once the layout of each screen was locked, a dedicated motion pass was run. The star-field drift, the clockwork orbital animation, the constellation line draw-ins, the nebula gradients, the meteor shower effects. Each one added through a focused prompt rather than a timeline or a code editor. The motion system is what makes Vesper feel alive, and all of it came through Stitch. Platform Feedback: Stitch's streaming generation is the most effective tool for dark, data-dense dashboard screens. There is no blank-canvas problem when a complex layout streams in live. The in-place editing workflow is where the real value lives: clicking a single element and describing a change in natural language ("make this feel more like a brass instrument") is closer to having a design conversation than using a tool. Here is the Project Live Link: https://vespersky-companion.netlify.app/

Pure sophistication. The contrast, the typography, the data layout and absolutely flawless execution.

Had a fun weekend building Merisole ☀️ Designed with Google Stitch.

I'm a senior product designer working on e-commerce sites every day, so my real question going into this challenge was practical: how easily does Stitch slot into my actual workflow? Could I use it to spin up a polished, client-ready option in a single sitting or even take it all the way to a shippable site? To test that, I asked AI to create a fictional footwear brand, Merisole, and treated it like a real client project from brief to launch.

How Stitch fit into the workflow Before generating anything, I wrote a custom design.md. Type scale, color tokens, spacing rhythm, the editorial tone I wanted Merisole to carry and uploaded it into Stitch as a design system. That single step is what made the rest of the build feel coherent. Every screen Stitch streamed onto the canvas already spoke the brand's language, so I wasn't fighting drift between sections. This is exactly the part of my normal e-com workflow that usually eats hours! Getting AI tools to respect a brand spec was something that Stitch handled cleanly. By the way, the stream on canvas feature was one of the coolest interactions I've seen so far! It was like having a designer design on your screen remotely through AnyDesk!

From there I worked mostly through in-place AI edits: clicking a section, telling Stitch "to add it to another variation," or "add a hover animation on the cards," and getting changes in seconds. The streaming generations meant I was reacting to a real interface taking shape instead of waiting for a static result.

The motion work is where Stitch surprised me most. Native hover states, shader backgrounds, and scroll-linked animation. All authored inside the tool, no handoff to a separate motion app, no code wrestling. Then I exported the build out to Netlify in one step.

New features I leaned on

Custom design.md - design system applied across the whole project (kept every screen on-brand from the first generation)

Streaming generations to canvas (the whole first pass)

In-place AI edits via prompts + point-and-click (most of the polish phase)

Native motion, hover states, and shaders on HTML canvas

Netlify export (final deploy)

Feedback on the platform Coming at this from a daily e-com design seat: Stitch genuinely shortens the distance between brief and client-ready. The design.md - design system flow is the underrated hero, uploading a brand spec once and having Stitch honor it across every streamed generation is the closest I've felt to "AI that respects a brand." The streaming canvas is amazing to say the least! It collapses the "describe - wait - judge - re-prompt" loop into something closer to direct manipulation. In-place edits felt the most like real design work. I stopped writing long prompts and just pointed at things. Two things I'd love next: An option to tweak elements without having to prompt it. And a way to save in-place edit "recipes" so I can reapply a vibe across sections without re-describing it.

Stitch didn't feel like an AI tool I was wrangling. It felt like a canvas that happened to understand what I meant. Merisole came together in a single sitting because of that, and I'm walking away convinced Stitch has a real place in how I'll pitch and build e-commerce work going forward.

Check it out:

Live site (really fun to click around): https://merisole.netlify.app

Stitch project: https://stitch.withgoogle.com/projects/834230787052196219

Amazing!