• 

• 

• 

• 

• 

• 

AuraStylist: Redefining Personal Fashion with Amazon Nova Inspiration

The fashion industry is shifting from generic trends to hyper-personalized expression. However, most styling apps either lack visual depth or fail to bridge the gap between inspiration and actionable shopping. We were inspired to build AuraStylist — a premium AI fashion house that does not simply suggest clothing, but understands a user's physical profile and aesthetic preferences through multimodal intelligence.

What it does

AuraStylist works as a private digital stylist. It analyzes a user's photo to understand:

• skin undertones
  
• body proportions
  
• face shapes
  

It then curates fashion recommendations for specific occasions and venues.

Users can preview these recommendations through AI-generated outfit visuals and directly explore similar products through a multimodal search pipeline that combines visual and textual understanding.

How we built it

AuraStylist is designed as a full-stack platform using:

• Next.js for the premium front-end experience
  
• FastAPI for backend orchestration
  

The complete AI pipeline runs on the Amazon Nova family through AWS Bedrock.

The Nova AI Engine

Nova Vision (Omni / Lite)

Acts as the visual analysis layer.

It performs:

• consultation photo analysis
  
• physical feature extraction
  
• reference image understanding
  
• style mood detection
  

Nova Reasoning (Pro / Omni)

Acts as the styling intelligence layer.

It processes:

• user preferences
  
• physical profile
  
• aesthetic references
  

And generates:

• styling recommendations
  
• color palettes
  
• garment suggestions
  

Nova Canvas

Acts as the visual generation layer.

It creates:

• realistic outfit previews
  
• style variations
  
• editorial-quality fashion outputs
  

Nova Multimodal Embeddings

Acts as the retrieval bridge between generated fashion ideas and searchable products.

It converts:

• outfit images
  
• style descriptions
  

into vector representations for semantic retrieval.

High-Precision Shopping Pipeline

AuraStylist uses a multi-layered retrieval architecture:

1. Attribute Parsing

Nova Vision separates generated looks into:

• Top
  
• Bottom
  
• Shoes
  
• Accessories
  

2. Multimodal Vectorization

Embeddings combine visual and textual context such as:

• fabric type
  
• color tone
  
• garment category
  

3. Vector Retrieval

Integrated with Amazon OpenSearch for similarity search across fashion catalogs.

4. Style Ranking

A ranking layer prioritizes products based on stylistic closeness to the generated recommendation.

Challenges we ran into

A major challenge was maintaining visual-text alignment.

Certain aesthetics such as Cyberpunk Minimalist are difficult to define using keywords alone.

This was improved by using multimodal reasoning, where image references and text prompts were interpreted together.

Another major focus was refining prompt quality so generated outfit visuals stayed visually consistent and premium.

Accomplishments that we're proud of

• Physical-profile-based personalization beyond simple filters
  
• Fast transition from generated concept to searchable products
  
• A luxury-style interface that feels product-ready
  

What we learned

Building with the Amazon Nova ecosystem showed the strength of unified multimodal systems.

Instead of combining unrelated vision and text pipelines, a single family of models improved consistency across:

• analysis
  
• reasoning
  
• generation
  
• retrieval
  

We also gained deeper practical understanding of vector search systems using OpenSearch.

What's next for AuraStylist

Virtual Try-On 2.0

Using inpainting for direct outfit application on user photos.

Contextual Wardrobe

Allowing users to upload their own wardrobe and receive mix-and-match suggestions.

Social Style Layer

A future feature where users can share style profiles and curated galleries.

Built With

• amazon-nova-canvas
• amazon-nova-lite
• amazon-nova-multimodal-embeddings
• amazon-nova-omni
• amazon-nova-pro
• aws-bedrock
• aws-opensearch
• boto3
• fastapi
• javascript
• lucide-react
• mongodb
• next.js
• pillow
• python
• tailwind-css
• uvicorn