Research & Technology

The science behind every GlamAR experience

GlamAR is built on purpose-trained AI - not off-the-shelf models. Every experience we ship is grounded in proprietary data, rigorous annotation, and continuous model improvement.

GlamAR research interface shield illustration
AI Facial Skin Analysis

Our approach to skin intelligence

Accurate skin analysis starts long before a model is trained. Our dedicated ML engineering team and AI researchers build from the ground up - starting with data quality, not model architecture.

Built on dermatology-informed annotation

Our dedicated annotation team labels skin images across diverse skin tones, ages, lighting conditions, and concern types. Acne, pigmentation, wrinkles, texture, dark circles - every concern is labeled with clinical precision before a single data point enters training.

The result is a model trained on over 5 million annotated images that performs consistently across skin types and real-world conditions - not just controlled environments.

Expert-labeled, pixel-verified

Every model update is measured against ground truth from trained annotators. Same image, three views: raw input, what an expert sees, what our model predicts.

Input skin analysis image
Input Image
Input skin analysis image with annotation review
Input Image
Input skin analysis image with model review
Input Image
  • Acne
  • Scars
  • Whiteheads
  • Enlarged Pores
Input skin analysis image
Ground truth skin annotation image
Model prediction skin analysis image
  • Acne
  • Scars
  • Whiteheads
  • Enlarged Pores
Original input skin concern image
Acne prediction skin concern image
Scars prediction skin concern image
Whiteheads prediction skin concern image
Pores prediction skin concern image
  • Acne
  • Scars
  • Whiteheads
  • Enlarged Pores
Virtual Try-on

Purpose-built models for every try-on category

Generic face meshes and off-the-shelf trackers break down in commerce. We train a separate model for each product category - optimized for real-time video, varied lighting, and the specific geometry each product demands.

Our ML engineers and AI researchers train category-specific models, optimized for video, not just static images, so tracking holds as the user moves in real time.

Eyewear

Face Tracking

Custom face landmark models trained for accurate frame placement across facial geometries - nose bridge, temple positioning, and scale.

Jewelry & Watches

Hand & Wrist Tracking

Real-time depth estimation and hand models trained for accurate ring, bracelet, and watch placement across hand shapes and skin tones.

Makeup

Lip & Eye Segmentation

Separate segmentation models per product category, trained to handle movement, shadows, and varied lighting conditions.

How we stay accurate

Research is not a one-time event

Our annotation team is not a historical input - they are an ongoing part of our model development cycle. As new products, categories, and use cases come online, the dataset grows. Models are continuously recalibrated against real-world performance data so accuracy improves over time.

1

Annotate

Label new data across categories

2

Train

Update models with expanded datasets

3

Deploy

Ship to production environments

4

Measure

Calibrate against real-world data

Schedule a call with our team

Trusted by global brands

FoxtaleLow Cost Glasses

Backed by enterprise grade security and scale

SOC 2GDPRISO certified