iOS

GlowTira: AI Glow-Up & Style Coach

Daily AI beauty & style coach with a calibrated face-scoring engine, identity-preserving try-on, and a proactive AI coach

App Store
Multi-pass
Scoring Engine
8+
Try-On Modalities
4.9 / 5.0
App Store Rating
Locked
Identity Fidelity

Project Overview

GlowTira is a daily AI beauty and style coach. Users upload a selfie and get an honest, calibrated read on their look, preview new hairstyles, makeup, beard styles, nails, tattoos, and piercings that still look unmistakably like them, and receive proactive, personalized coaching to actually act on the advice. At the core is the GlowScore engine: a multi-pass GPT vision pipeline that analyzes symmetry, skin clarity, the eye area, lips, jawline, and overall harmony, then calibrates and anchors those scores to a locked personal baseline so progress is measured fairly over time rather than reset on every scan. On top of that sits Lumi, an AI style coach with long-term memory that nudges users with daily proactive messages, challenges, and milestones. The project covered the full product lifecycle: designing and building the React Native experience, architecting a Supabase backend with 28 Edge Functions orchestrating GPT and fal.ai FLUX, engineering the calibrated scoring and coaching systems, localizing into 7 languages, and shipping a freemium subscription model on the App Store.

The Challenge

Building a credible AI glow-up product is deceptively hard. It only works if the scores feel honest, the try-on results are unmistakably *you*, and the coaching feels genuinely personal. Several core challenges had to be solved: **1. Trustworthy, Non-Gameable Scoring** A single GPT vision call returns inconsistent, easily-gamed numbers. The score had to feel insightful and fair, stay stable across lighting and angles, and reward real progress, not better selfies. This demanded a multi-pass engine with calibration, confidence bands, and smoothing rather than a naive prompt. **2. Identity-Preserving Generation** The biggest risk in AI beauty apps is the "uncanny" result, where a new hairstyle or makeup look no longer looks like the user. Preserving facial identity while convincingly altering hair, beard, makeup, and accessories required the right model per modality and careful conditioning. **3. Many Try-On Modalities, One Pipeline** GlowTira spans hairstyles, beards, makeup, nails, glasses, bandanas, tattoos, and piercings, each with different visual rules. The system needed one flexible generation pipeline routing every modality to the right fal.ai model and parameters without fragmenting the codebase. **4. A Coach That Remembers** Generic chatbot replies feel hollow. Lumi needed long-term memory of each user's routine, style profile, goals, and history, plus proactive daily outreach, threaded topics, and a sense of an evolving relationship, all while keeping token costs sane. **5. Latency, Cost & Quality at Scale** High-quality generation and multi-pass vision are slow and expensive. Balancing render quality against wait times, while controlling per-call cost through caching, gating, and async processing, was essential to both retention and unit economics. **6. Privacy & Sensitive Imagery** The app processes users' personal face and body photos. Secure handling, scoped storage, and deletion, with clear consent, was non-negotiable.

The Solution

I designed and built GlowTira end to end, combining a polished React Native client with a Supabase backend purpose-built for orchestrating AI: **React Native + Expo Client** Built the iOS app with React Native 0.81 and Expo SDK 54 (Expo Router) for fast iteration and a single, maintainable codebase. Used Zustand and React Query (with an AsyncStorage persister) for state and caching, and a warm sage-and-champagne design system with glassmorphism for a calm, premium feel. **GlowScore Engine** Engineered a multi-pass scoring pipeline: GPT vision extracts descriptors, raw per-dimension scores are calibrated for anti-gaming and anchored to a locked baseline, confidence bands gate low-quality photos, and EWMA smoothing across scans keeps results stable. Every score returns clear, encouraging, deduplicated actions instead of raw numbers. **Identity-Preserving Try-On** Routed each modality through the right model: FLUX.1 Kontext for face-aware hairstyle and beard generation that preserves identity, and FLUX.2 for outfit and accessory swaps. One unified generation layer powers hair, makeup, nails, beard, glasses, bandanas, tattoos, and piercings. **Lumi AI Coach** Built a coaching system on GPT-driven reasoning with persistent per-user memory (routine, style profile, goals, trust score, relationship stage), threaded conversations by topic, and scheduled proactive daily messages, challenges, and milestones via Supabase Edge Functions. **Supabase Backend** Architected a Postgres schema (60+ migrations) with Auth, Storage buckets for scans and generations, and 28 Edge Functions handling face/outfit analysis, generation, coaching, diary insights, trend reports, and push notifications. All third-party keys live as Supabase secrets, never exposed to the client; long renders use async polling plus push. **Monetization & Privacy** Integrated RevenueCat and Superwall for entitlements, paywalls, and receipts, gating unlimited scans and premium try-ons. Implemented consent-first onboarding with scoped storage and deletion controls for user photos.

Development Approach

1

Discovery & Concept: Researched the glow-up market, defined the "honest score + preview + coach" product loop, and mapped every try-on modality and coaching surface into one vision

2

Design Phase: Crafted a calm, premium design system (warm ivory, sage, champagne, glassmorphism) with onboarding and an effortless capture-to-result flow

3

Scoring R&D: Engineered the GlowScore engine, moving from a naive single GPT call to a multi-pass pipeline with calibration, confidence bands, baseline anchoring, and EWMA smoothing

4

Generation Pipeline: Built the unified try-on layer routing each modality to the right fal.ai model (FLUX.1 Kontext for face-aware hair/beard, FLUX.2 for outfit/accessory swaps)

5

AI Coach: Developed Lumi with persistent per-user memory, threaded topics, and scheduled proactive messages, challenges, and milestones

6

Backend & Scaling: Implemented the Supabase schema, Storage, and 28 Edge Functions with async polling, caching, and tier-based gating for reliable AI workloads

7

Monetization, i18n & Privacy: Integrated RevenueCat and Superwall, localized into 7 languages, and shipped consent-first media handling with deletion controls

8

Launch & Iteration: Submitted to the App Store, monitored scoring quality and funnel analytics, and iterated on prompts, pricing, and UX based on real usage

Technologies Used

Mobile Development

React Native 0.81TypeScriptExpo SDK 54Expo RouterReanimatedSora + Plus Jakarta Sans

State Management & Data

ZustandReact QueryAsyncStorage Persister

Backend & Infrastructure

SupabasePostgresSupabase AuthStorageEdge FunctionsPush Notifications

AI & Generative Imaging

GPT VisionGlowScore Enginefal.ai FLUX.1 Kontextfal.ai FLUX.2Identity-Preserving Generation

Monetization & Growth

RevenueCatSuperwallApp Store In-App PurchasesPaywalls

Localization & Analytics

i18n (7 languages)Custom Supabase AnalyticsTrend Reports

Key Features

✨

GlowScore Engine

Multi-pass GPT vision scoring of symmetry, skin, eye area, lips, jawline, and harmony, calibrated and anchored to a locked baseline so progress is measured fairly.

💇

Hairstyle & Beard Try-On

Browse 50+ styles and generate face-aware previews with FLUX.1 Kontext that keep your identity intact.

💄

Makeup & Nails

Preview makeup looks and nail designs in seconds, from natural to bold, all rendered to still look like you.

🎨

Tattoo, Piercing & Glasses

Visualize tattoos on any body region, piercings, frames, and headwear before committing.

👗

Outfit & Date Prep

AI feedback on color match, coherence, and proportions, plus occasion-aware date-prep planning.

🧠

Lumi AI Coach

A coach with long-term memory that sends proactive daily messages, challenges, and milestones tailored to your routine and goals.

📔

Diary & Insights

Log daily mood, skin, and products, then get weekly AI-powered trend insights and a suggested focus.

📈

Progress Tracking

Daily streaks, weekly and monthly reports, and trend sparklines compared against your baseline.

Results & Impact

Multi-pass
Scoring Engine
GlowScore: GPT vision with calibration, confidence bands, and EWMA smoothing
8+
Try-On Modalities
Hair, beard, makeup, nails, glasses, bandanas, tattoos, and piercings in one app
4.9 / 5.0
App Store Rating
Strong early ratings driven by honest scores and identity-preserving results
Locked
Identity Fidelity
fal.ai FLUX keeps previews recognizably the user across every modality
28
Edge Functions
Supabase functions powering analysis, generation, coaching, and reports
7
Languages
Fully localized: EN, TR, AR, FR, IT, ES, DE
Daily
AI Coach
Lumi remembers each user and reaches out proactively with tailored guidance
iOS
Platform
Native experience built with React Native and Expo
"
Finally a glow-up app that feels honest. The score actually explains why and gives me three things to work on instead of just a number. I tried like 15 hairstyles in five minutes and every one still looked like me, not some random model. The daily coach messages are weirdly motivating. Obsessed.
S
Sophia M.
Beauty Enthusiast, App Store Review

Related Articles

Mobile Development

Mobile App Monetization Strategies 2025: Complete Revenue Guide

Master proven monetization models from freemium to subscriptions. Learn pricing psychology, conversion optimization, and revenue strategies that generate sustainable income from iOS and Android apps in 2025.

Have a Similar Project in Mind?

Let's discuss how I can help bring your mobile or web application idea to life with cutting-edge technology and exceptional user experience.

Get in Touch