AI Hairstyle Recommender
AI hairstyle recommendation using facial feature analysis
Agency
Nep Tech Pal Pvt. Ltd.
Category
android
Type
android
Status
published
Key Features
Discover the powerful features that make this project stand out.
Face Shape Analysis
AI detects your face shape (oval, round, square, heart, oblong) and recommends matching styles.
200+ Hairstyle Library
Curated library of hairstyles for men and women across all face shapes and hair types.
Virtual Try-On
Realistically apply any hairstyle to your photo using AI image generation.
Virtual Hairstyle Try-On
AI overlay shows any hairstyle on your actual face in real time — see before you commit to a cut at the salon.
Before & After View
Side-by-side comparison of your current hairstyle vs the recommended style.
Recommendation Engine
Personalized hairstyle recommendations based on face shape, existing hair texture, and style preferences — backed by professional styling principles.
Salon Sharing
Export your chosen hairstyle as a reference card to show your hairdresser.
Salon Finder Integration
Find nearby salons that specialize in your recommended style, with ratings and booking links — connecting the discovery to action.
Hair Color Preview
Try different hair colors alongside hairstyle changes simultaneously.
Before/After Comparison
Side-by-side comparison of your current look vs. any virtual try-on — easy to share with your stylist or friends for opinions.
From Challenge to Solution
Discover how we transformed challenges into innovative solutions.
The Challenge
The Challenge: Turning a Stylist's Trained Eye into an AI Algorithm
Getting a new haircut is one of the most anxiety-inducing decisions many people make — and for good reason. A haircut takes months to grow back from, and the wrong one can significantly affect how you feel about your appearance. Professional stylists spend years learning to read face shapes and match them to complementary hairstyles. Encoding this expertise into an AI system that works accurately across diverse face types was the central challenge.
Consider how Warby Parker solved virtual glasses try-on: they needed AR face tracking accurate enough that users could trust the virtual try-on to reflect reality. Any inaccuracy — a frame that appeared to fit perfectly virtually but looked wrong in person — would erode trust instantly. The same trust bar applies to hairstyle try-on: users must be able to rely on the virtual preview as a genuine indicator of how the style will look.
The specific challenges:
- Accurate face shape classification: Face shape is a spectrum — no face is perfectly oval or perfectly square. The classification model must produce confident, accurate categorizations from real-world photos with varying lighting, angles, and facial hair.
- Diverse hair texture representation: Hairstyle recommendations that work for straight Asian hair may look completely different on South Asian or African hair types. The recommendation engine needed to account for hair texture, not just face shape.
- Realistic try-on overlay: Placing a hairstyle photo overlay on a user's face requires accurate head segmentation and hair mask blending. Obvious composite artifacts (visible edges, wrong scale, wrong angle) make the try-on feel useless.
- Hairstyle library curation: 200+ hairstyle reference images had to be photographed or sourced consistently — same background, consistent framing — for the overlay system to work correctly.
- Cultural relevance: The primary market includes South Asian users whose hairstyle preferences and cultural context differ significantly from Western styling references. The library and recommendations needed to reflect this.
Our Solution
Solution: MediaPipe Face Mesh + Professional Styling Knowledge Graph
We combined computer vision precision (MediaPipe's 468-point face mesh) with structured hairstyling domain knowledge to create a recommendation system that outperforms generic style apps by being tuned specifically to real styling principles.
1. High-Accuracy Face Shape Classification
MediaPipe Face Mesh provides 468 3D facial landmarks from a single photo. We extract 12 key measurements: face width at forehead, cheekbones, and jaw; face length; jawline angle; and forehead width-to-height ratio. A rule-based classifier using these proportions assigns a primary face shape with a confidence score. When confidence is below 80% (borderline face shapes), the system presents recommendations for both closest matches. Tested across 500 diverse reference photos, classification accuracy reached 91% — comparable to the accuracy of professional stylist assessments in blind studies.
2. Recommendation Engine with Styling Principles
We consulted with three professional hairstylists to encode their decision-making principles into a recommendation matrix: which cuts elongate round faces, which styles add width to narrow faces, which lengths balance strong jaw lines. The engine weighs face shape (primary factor), hair texture (fine/medium/thick, straight/wavy/curly), and user style preferences (low maintenance vs. styled). The result is a ranked shortlist of recommendations with plain-English explanations — "This layered bob will soften your square jaw while adding volume at the crown to balance your face shape." This explainability is what separates a trusted recommendation from a black-box suggestion.
3. Virtual Try-On Overlay System
The try-on overlay uses head segmentation (MediaPipe's selfie segmentation model) to create a precise mask of the user's hair region. Each hairstyle reference image is pre-processed to extract just the hair portion on a transparent background. At try-on time, the hairstyle layer is scaled to the user's head width, aligned to the hairline landmarks, and composited using alpha blending. Users can adjust scale and position manually for fine-tuning. The result is convincing enough that users regularly share their try-on screenshots with their stylist as a direct reference — which is exactly the salon-bridge use case we designed for.
4. Salon Finder Integration
After discovering their ideal hairstyle, users are shown nearby salons that have been tagged as specializing in the relevant cut style. Salons are sourced from Google Places API and filtered by user reviews mentioning relevant style terms. This end-to-end journey — discover → try on → find a salon — converts the app from a novelty into a practical tool that drives real-world appointments.
Technology Stack
The powerful technologies used to bring this project to life.
FastAPI
Backend
MediaPipe
Backend
Python
Backend
Stable Diffusion
Backend
Firebase
DevOps
Dart
Mobile
Flutter
Mobile