Mira AI Stylist
An AI stylist for the clothes you already own. It reads your weather, calendar and closet, then hands you one look for today.
Getting dressed is a tiny decision you make every single morning.
Mira started in 2023 with one question: could generative AI help people get more out of the clothes already in their closet? Not buy more. Wear better. Most styling apps showed inspiration from strangers with different bodies and different weather. Mira starts from your day instead.
Three designers, one brief. I owned how it looks.
We ran research, flows and testing together as a team. The visual side was mine.
Built the visual system
Color, type and the frosted card style, tuned so the outfit is always the loudest thing on screen.
Directed the AI imagery
AI images drift between generations. I kept one character consistent across every screen so the app reads as one person's day.
Told the story
Designed the presentation deck, this case study and the 2026 award entry.
Ten weeks, from survey to high fidelity.
Research first, pixels last. Each step fed the next one.
Survey, interviews, competitors
58 survey responses, 9 interviews, and a teardown of Weather Fit, XZ, StyleDNA and Dada (搭搭).
Personas and journey maps
Two personas and two journey maps: getting dressed for a normal day, and packing for a trip.
Features roadmap and IA
4 key flows and 10 features, mapped into an information architecture for the whole app.
Wireframes, then high fidelity
Low fidelity flows first, then five rounds on the home screen until it worked.
Usability tests
Tested with 5 people. Card and mood board layouts beat list layouts.
Most people care about their outfits. Few have time to.
37 of 58 people, 63.8%, rated how much they care about outfits above 6 out of 10. The same people told us mornings are rushed, the forecast doesn't match how warm they actually feel, and work events raise the stakes.
None of them fit the outfit to the person: body shape, how warm or cold they run, or the fabrics they avoid.
Nobody said they hated fashion. They hated the clock.
A fashion design student, an interior designer, a marketing manager, an executive director. Even the people who loved picking clothes described the same squeeze.
Weather uncertainty
A forecast doesn't tell you what 12°C feels like on your commute.
Decision fatigue
Busy mornings leave no time to try things on, then change again.
Special occasions
Work, formal events and anything that ends up in photos raise the stakes.
Repetitive outfits
People lose track of what they wore, so the same few looks come back.
Lack of confidence
Online inspiration rarely matches your body, so you can't tell if it will work.
No style feedback
Nothing remembers what you liked, so every morning starts from zero.
Meet Cecilia, who has nothing to wear in a full closet.
Cecilia is a fashion designer in New York. She has meetings and fittings all day, and she feels judged on every outfit. We mapped her morning minute by minute, from 8:00 when she starts thinking about clothes to 23:00 when she is still second guessing them.
The lowest points sit at trying things on and waiting for feedback. That's where Mira needed to help most.
Four flows, ten features.
We grouped every opportunity from the journey map into four places in the app: Homepage, AI Stylist, Closet and Settings. Then we drew the information architecture so each feature had one clear home.
The AI Stylist flow, in grey boxes first.
The stylist flow had the most branches: know your style or skip, pick from four outfits, regenerate, try it on the mirror, rate it. We drew the decision tree before any pixels.
Each flow answers specific pain points.
Every feature traces back to a pain point from the research.
See the look on you before you change, and get a forecast tuned to whether you run warm or cold.
Ideas based on your settings, plus a rating and comments step that builds confidence over time.
A closet sorted for you, outfits built around a single item, and a calendar of what you wore.
Five versions before the home screen felt like a mirror.
The home screen carries the whole product: weather, schedule, the outfit and the actions. Each version fixed one thing the last one got wrong.
Illustrated avatars
Easy to draw, but nobody pictures themselves as a cartoon. The day's timeline showed up in V1 and stayed.
Realistic AI people
The outfit suddenly read as something you could wear. The schedule shrank to one event marker.
The whole day as a list
Client meeting, party, movie night. Useful, but the panel covered most of the outfit.
Change the look
Tap the one piece you don't like and swap only that. The yellow Regenerate button became the one bright thing on screen.
Every flow, end to end.
The final prototype covers the full AI Stylist journey, the closet, and settings like body measurements, thermal sensation and style preferences.
Same person, different city, different day.
22°C in Seattle with a movie night gets a soft knit and joggers. 12°C and rain in New York before a client meeting gets a trench, dark denim and boots. One tap regenerates. Try on mirror carries the look to the mirror.
Why this look? Trust needs a reason.
Three years later, AI stopped waiting to be asked. That made the old Mira feel incomplete. If an app tells you what to wear without saying why, it's hard to trust and harder to correct. So each suggestion now shows the weather, the event and the style preference behind it. Try it.
If you can see why the AI picked it, you can fix the part it got wrong.
Design principle for Mira 2.0- WeatherPartly cloudy, 22°C. Light layers.
- CalendarClient meeting at noon. Polished but easy.
- StyleYour formula: soft neutrals, relaxed fit.
A low fidelity concept added for the 2026 version. Not tested with users yet.
One stylist across your phone, a widget and the mirror.
The app is for deep control: closet, preferences, body measurements. A widget shows today's look at a glance. The mirror shows the outfit on you before you put it on. Mira suggests, you decide.
Calm enough to use half awake.
People open Mira in the first ten minutes of their day. I kept the palette cool and soft, gave the product one warm color, and saved it for the actions that move you forward.
#101836Type and icons. Softer than black.
#E6C66BOnly for Regenerate and Style the look.
#A9B1CCRainy day gradients and panels.
#EEF1F8Backgrounds and frosted cards.
A student sprint that held up three years later.
MUSE Silver 2026
Product Design: UX / UI / IxD, entered as Harvard University × University of Washington.
Cards over lists
In usability tests with 5 people, card and mood board layouts beat list layouts.
App to ambient
The 2026 revisit added reasoning and a multi surface concept on top of the original app.
What I'd tell myself in 2023.
Polish can hide gaps
AI imagery made Mira look finished on day one. A beautiful screen gets fewer hard questions, and the hardest one, why should I trust this, waited three years for an answer.
Test the reasoning next
Why this look is still a concept. Next I'd put it in front of people and check whether reasons build trust or just make the screen busier.