Fashionnd.
Most styling apps style outfits. Fashionnd styles you — your body, your color, your closet, your life.
The Hannah Montana closet, reimagined for your body, color, and life.
Most styling software flattens "what should I wear" into a feed of products and a leaderboard of trends. Fashionnd proposes the opposite — a unified, private surface built on what you already own, what your body actually is, and what colors your face actually belongs to.
Seven layers — body scan, color analysis, Style DNA, closet, outfit-of-the-day, virtual try-on, and shop-the-gap — feed each other in one direction: every signal sharpens the next recommendation. The output isn't more shopping. It's better dressing, with reasons.
One profile, seven systems.
Body Scan
ARKit + MediaPipe pose estimation captures measurements and posture from a slow turn — rounded shoulders, forward head, real proportions. Falls back to photogrammetry on Android. Data stays on-device unless the user opts to sync.
Color Analysis
A guided selfie under a calibration step (white paper / daylight) classifies the user's seasonal palette across 12 sub-seasons. Detection runs on the Monk Skin Tone Scale plus a custom contrast model. The palette becomes a filter every downstream system uses.
Style DNA
Body, color, Kibbe archetype, and a lifestyle survey distilled into one identity surface — with a parametric Ready-Player-Me-style avatar shaped to real measurements. Editable: the model learns from corrections. The 'mood' line is LLM-generated.
The Closet
Photograph an item once and a CLIP-based embedding model fills in 14 attributes — fabric, color, category, formality, season. Email parser pulls Shopify and Amazon order receipts. Wear counts surface neglected pieces and make cost-per-wear visible.
Outfit of the Day
An outfit composed from the closet, weather, calendar, palette, and wear-history — with a written rationale. The 'because' is what turns a recommendation into trust. Cycles intelligently; never repeats yesterday.
Virtual Try-On
Tap any item — owned or for-sale — onto the avatar. Three.js cloth simulation runs on-device for owned items; OOTDiffusion / IDM-VTON renders photoreal try-ons server-side. Swap colors and silhouettes side-by-side before checkout.
Shop the Gap
The opposite of fast fashion. Capsule analysis identifies the smallest set of items that would unlock the most new outfits, with transparent match scores (body + color + lifestyle + closet fit). Affiliate revenue, never push-driven.
From a slow turn to tomorrow's outfit.
Each layer is a measurement; each measurement narrows the next recommendation. The model is the product.
- Step 01
Capture identity
Body scan + color analysis + lifestyle survey produce a structured profile in the first session. Three minutes of capture, on-device.
- Step 02
Build the closet
Photo-to-catalog auto-tags every garment via a CLIP-based embedding model. Email scrapers backfill from order confirmations.
- Step 03
Compose & explain
Claude reasons over closet embeddings, weather, calendar, and the user's palette to produce daily outfits with rationale and cycle logic.
- Step 04
Try, decide, fill
Three.js + cloth physics for owned try-on, OOTDiffusion for shop items. Capsule analysis identifies wardrobe gaps and surfaces only what unlocks the most new looks.
Eight screens, one click each.
Welcome → body scan → color analysis → Style DNA → closet → outfit-of-the-day → virtual try-on → shop the gap. Each screen has designer notes and the technical decisions behind it.
Open the prototype↗Pose estimation, embeddings, and reasoning.
- ●React Native
- ●Expo
- ●Lottie
- ●ARKit
- ●MediaPipe
- ●Bodygram SDK
- ●Photogrammetry fallback
- ●CLIP embeddings
- ●Monk Skin Tone Scale
- ●OOTDiffusion / IDM-VTON
- ●Custom contrast model
- ●Claude API (rationale + mood)
- ●Custom outfit-scoring model
- ●Three.js
- ●Cloth physics
- ●Ready-Player-Me-style avatar
- ●Shopify Storefront API
- ●Nordstrom + ASOS APIs
- ●Email receipt parser
"The next generation of fashion software won't sell us more clothes. It'll quietly make us better at the ones we already own."
— Project thesis, Fashionnd