A closet app that shows value from the first photo.

I wanted a closet app I'd actually open at 6 a.m. I started out trying to make adding clothes faster.

Research changed the goal. Cataloging a whole closet takes time however it's done, because every piece has to be photographed. The apps I tested made you finish that chore before you could build a single outfit.

So I stopped trying to win on speed alone. The app gives you a look from the first photo, and the closet fills itself in the background while you focus on what to wear.

I timed the three closest competitors myself.

What I found

• Slow to value. 6 to 8+ minutes before a first saved outfit, onboarding included.
• Two jobs, never one. Every app could save a look or catalog its pieces, but never both at the same time.
• Your photo gets lost. Alta keeps an avatar instead of your photo and locks items in place on its canvas. Acloset drops the photo when it splits out pieces.
• Busy screens. Alta and Acloset pile on options, so even browsing takes effort.
• No duplicate check. None of the three noticed the same jeans added twice.

I have looked at apps like that, but they always seemed like more trouble than they were worth.

A friend, on closet apps

The bet. If one photo saves the outfit and catalogs its pieces, a busy person can build a closet without ever sitting down to set one up.

• Who: busy people who like staying organized but don't have time for setup
• Success: time from photo to saved outfit
• Trust: how many pieces people accept without fixing

Everything hinged on one moment: does a photo of me become pieces I'd recognize? Raw cutouts were jagged, and no option was good enough on quality, cost, speed and privacy all at once.

So I layered them. Free on-device cutouts show up instantly, and a cleaner cloud version replaces each one in the background. Saving never waits on quality.

A bar I could defend. I didn't need perfect detection. The bar: every main piece is recognizable, and anything that's off takes one tap to fix. No AI gets every piece right, so a fast fix earns more trust than chasing perfection.

Capping cost before it could surprise me. Each cloud cleanup costs about 4¢ a piece. Uncapped, one big closet or one bug could run up a real bill. So I capped it at $5 a month for the whole app and 30 cleanups a day per phone. The tradeoff is that some pieces stay rougher for a while, but the bill can never surprise me. With thousands of users, the same cap becomes a pricing question: what's free, and what's paid.

I treated AI like a team I was leading. I set the goal and the plan, each tool got one job, and every decision stayed with me.

It started with boundaries. Every agent got the same rule, and nothing merged without my review.

Only change homepage files. Don't modify shared components. If you think one needs changing, tell me instead.

My prompt to a parallel agent

Then I matched the tool to the decision. I explored ideas in Claude Design and switched to Figma when I already knew what I wanted, since doing it myself was faster than describing it.

When I did hand design work to AI, I art-directed it the way I would a teammate. For the pose guide, I described the animation as beats (step back, arms out, hold still, pieces lift off), and it landed in one prompt. Then I cut it: a whole screen before the camera slowed down the one thing people came to do. It became a line of rotating tips.

I also learned where speed was safe. Interface work is cheap to change, so I built it fast and fixed it as I went. Anything that touched people's data, I planned first. Everything lives on the phone, so clearing Safari could wipe a closet. Before any code, I had Cursor compare options on protection, effort, cost and user impact, then pushed on scale: who pays to store other people's photos? I chose a backup file for now, with cloud sync once real users need it.

PLAN ONLY — do not write code. […] For each option: what it protects against, effort, cost, and what changes for me as the user. Recommend one.

…am I the one paying for all those photos getting stored by others?

My prompts to Cursor, before building backup

When something broke, I went after the cause instead of the next patch. Cleanup once failed silently while the agents kept offering fixes. I understood the system well enough to ask the question that mattered: had any paid cleanup actually run? None had, which led straight to a one-line fix instead of a pile of guesses.

Every detail below exists so people trust what the app did with their photo.

The moment it clicks. Tap Confirm and watch your pieces fly into the closet. That feedback tells you setup is happening without you.

A wait that explains itself. Cleanup takes a few seconds per piece. Meanwhile, each card shimmers in the shape of its piece and says what's happening, and you can keep going.

A duplicate check. None of the apps I tested noticed the same jeans added twice. Here, a likely match asks "Is it the same piece?"

A canvas that feels like play. Pieces land spread out, never stacked. Pinch to resize, twist to rotate, and tap an item to switch the toolbar to that item's tools.

Pulling back a feature. The On me / Items toggle failed its first phone test: switching to Items made my photo disappear with no way back. It was a hard call, but I pulled it. A look keeps your photo, and the pieces are one tap away in the closet, where you just watched them land. Next I'll test whether people would rather style with pieces than with their photo, and explore easy ways to do that.

Rules, not one-off screens. Wayfinding drifted as agents built: some screens closed with an X, some with Back, some with a pop-up, some with a transition. I wrote one set of rules for how you leave, go back and confirm, put it where every agent reads it, and rebuilt shared components to follow it. I held everything else to the same standard: when one screen's filter tags used the wrong text style, I didn't fix that screen, I made the tags a shared component.

When to use X vs. Back. X means starting something new; leaving without Save throws away the draft. Back means editing something that already exists; leaving keeps changes. Never put X and Back on the same screen.

From the app's rules file, which every agent reads on every task

It's a working iPhone app on a live link, built in about a month, part-time. It hasn't met real users yet, so the next step is finding out whether it works for someone who isn't me.

On my own phone, the outfit is saved in 3 minutes and the cleaned pieces land within a minute after, in the background. The apps I tested took 6 to 8+ minutes, onboarding included. One user proves the flow works, not that it works for everyone.

What I'd take to a team. The same habits scale: a clear brief that says what's out of scope, shared rules every contributor reads (person or agent), and a review gate before anything ships.

Testing next with 5 busy, organized people. It passes if:

• photo to saved outfit stays under 5 minutes
• at least 75% of pieces are accepted without fixes
• every main piece is recognizable on each person's five most-worn outfits
• people still choose to style with their own photo, not only pieces

After that: fix what slows people down, starting with cleanup speed; rework the camera guide; replace the backup button with automatic backup. Then the next bet: outfit ideas that know the weather and season ("62°F, rain at 3") and learn from what you pick.