How It Works

The Style Graph is Glimmer's core intelligence layer. It's a network of your style preferences that grows and sharpens every time you capture a snapshot.

The pipeline

When you capture a snapshot, Glimmer's AI runs a multi-step analysis:

StepWhat happens
Screenshot analysisAI examines the captured image and page URL
Product detectionIdentifies products, brand, and shop from the page
Prototype extractionBreaks each product into canonical style signals called prototypes
Graph connectionLinks prototypes that appear together, strengthening co-occurrence edges

What's a prototype?

A prototype is a normalized style signal. Instead of storing "Anine Bing Camel Linen Blazer," Glimmer decomposes it into reusable traits:

DimensionExample
Roleouterwear
Categoryblazer
Silhouetterelaxed
Colorcamel
Materiallinen

This normalization means a camel blazer from Anine Bing and one from Sandro reinforce the same style signals in your graph.

Co-occurrence: the graph's edges

When two prototypes appear together repeatedly — say "earth tones" and "relaxed silhouettes" — Glimmer records that co-occurrence. The more you capture items with both traits, the stronger that edge becomes.

These connections power recommendations. If your graph shows a strong link between "minimalist" and "neutral tones," Glimmer knows to surface products that match both.

The 10-snapshot threshold

Glimmer needs at least 10 snapshots to generate meaningful style insights. Before that, the graph is too sparse for reliable patterns. The drawer shows your progress toward this threshold.

After 10 snapshots, you unlock:

  • Style Brief generation
  • Accurate affinity scoring
  • Personalized search results

The more you capture, the more precise everything gets.

Your graph is personal

Every member has their own Style Graph. Your snapshots, prototypes, and co-occurrence data are scoped to your account and never shared with other users.