> ## Documentation Index
> Fetch the complete documentation index at: https://withlooksy.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# How Looksy Virtual Try-On Technology Works

> Learn about the AI technology powering Looksy virtual try-on. Understand how machine learning, computer vision, and rendering create realistic try-on experiences.

## The Technology Behind Virtual Try-On

Looksy uses state-of-the-art artificial intelligence to create realistic virtual try-on experiences. Here's how the technology works.

## AI Try-On Technology

Looksy is powered by **cutting-edge artificial intelligence and machine learning** that transforms how customers shop for apparel online. Our AI models deliver photorealistic virtual try-on results in 15-20 seconds.

### How AI Powers Virtual Try-On

**Core AI Technologies:**

<CardGroup cols={2}>
  <Card title="Convolutional Neural Networks (CNNs)" icon="brain">
    Deep learning models that analyze images, detect features, and understand garment structure
  </Card>

  <Card title="Generative Adversarial Networks (GANs)" icon="wand-magic-sparkles">
    AI systems that generate realistic garment renderings adapted to each customer's body
  </Card>

  <Card title="Computer Vision & Pose Estimation" icon="eye">
    Advanced algorithms that map body structure, detect poses, and understand spatial relationships
  </Card>

  <Card title="Style Transfer Algorithms" icon="palette">
    Machine learning techniques that maintain garment patterns, colors, and textures during rendering
  </Card>
</CardGroup>

### AI Model Training

**What makes Looksy's AI accurate:**

* **100+ million parameters:** Sophisticated neural networks capable of understanding complex garment-body relationships
* **Millions of training examples:** Models trained on diverse datasets of garments, body types, skin tones, and poses
* **Continuous learning:** AI improves over time as it processes more try-ons
* **Apparel-specific training:** Unlike generic AI, Looksy's models are trained exclusively on fashion products

<Note>
  Looksy's AI is continuously updated with new training data to improve accuracy, support more product types, and adapt to fashion trends.
</Note>

### AI vs. Traditional Virtual Try-On

**Why AI-powered try-on is superior:**

| Approach           | Traditional (AR Overlay)  | AI-Powered (Looksy)                 |
| ------------------ | ------------------------- | ----------------------------------- |
| **Realism**        | Basic overlay, looks fake | Photorealistic rendering            |
| **Adaptation**     | No body-specific fitting  | Adapts to each body type            |
| **Speed**          | Real-time but low quality | 15-20 seconds, high quality         |
| **Accuracy**       | 60-70%                    | 90-95%                              |
| **Garment Detail** | Limited                   | Preserves patterns, logos, textures |

AI-powered virtual try-on produces **realistic previews** that help customers make confident purchase decisions, not just basic overlays.

## High-Level Overview

<CardGroup cols={2}>
  <Card title="1. Image Analysis" icon="magnifying-glass">
    AI analyzes the product image and customer selfie
  </Card>

  <Card title="2. Body Mapping" icon="person">
    Computer vision identifies body structure and positioning
  </Card>

  <Card title="3. Garment Adaptation" icon="shirt">
    AI adapts the garment to match body shape and pose
  </Card>

  <Card title="4. Realistic Rendering" icon="wand-magic-sparkles">
    Advanced rendering creates a photorealistic composite
  </Card>
</CardGroup>

**Total time:** 15-20 seconds from upload to result

## Step-by-Step Process

### Step 1: Customer Uploads Selfie

**What happens:**

* Customer clicks "Try On"
* Takes photo or uploads from library
* Image is sent securely to Looksy servers

**Technical details:**

* Image is compressed for fast upload
* HTTPS encryption ensures security
* Original image is not stored after processing

### Step 2: Face and Body Detection

**What the AI does:**

* Identifies facial landmarks (eyes, nose, mouth)
* Detects body pose and positioning
* Measures proportions and dimensions
* Identifies skin tone and lighting conditions

**Technologies used:**

* Convolutional neural networks (CNNs)
* Facial landmark detection algorithms
* Pose estimation models
* Color analysis and tone mapping

**Output:** Body map with key points and measurements

### Step 3: Product Image Analysis

**What the AI does:**

* Isolates the garment from background
* Identifies garment type and structure
* Detects patterns, colors, and textures
* Maps neckline, sleeves, and hemline

**Technologies used:**

* Semantic segmentation
* Object detection
* Texture analysis
* Edge detection

**Output:** Garment mask and feature map

### Step 4: Garment Adaptation

**What the AI does:**

* Adapts garment to match body shape
* Adjusts proportions for realistic fit
* Maintains garment characteristics (patterns, logos)
* Applies perspective and depth

**Technologies used:**

* Generative adversarial networks (GANs)
* Style transfer algorithms
* 3D pose estimation
* Geometric transformations

**Output:** Adapted garment ready for compositing

### Step 5: Realistic Rendering

**What the AI does:**

* Blends garment with selfie
* Matches lighting conditions
* Applies shadows and highlights
* Ensures natural edges and transitions

**Technologies used:**

* Image synthesis
* Lighting estimation
* Shadow rendering
* Color matching

**Output:** Final photorealistic try-on result

<Check>
  The entire pipeline processes in 15-20 seconds on Looksy's servers, with no strain on your store or customer devices.
</Check>

## AI Models and Training

### Deep Learning Architecture

**Model types:**

* **Encoder-decoder networks:** For image segmentation
* **GANs (Generative Adversarial Networks):** For realistic garment rendering
* **CNNs (Convolutional Neural Networks):** For feature detection
* **Transformer models:** For attention mechanisms

**Model size:** 100+ million parameters

**Training data:**

* Millions of garment images
* Diverse body types and skin tones
* Various lighting conditions
* Multiple garment categories

<Note>
  Looksy's models are continuously updated with new data to improve accuracy and support more product types.
</Note>

### What Makes Looksy's AI Different

**Advantages over competitors:**

1. **Apparel-specific training:** Models trained exclusively on clothing (not generalized)
2. **Diverse datasets:** Inclusive training data across body types and skin tones
3. **Shopify-optimized:** Fine-tuned for e-commerce product images
4. **Fast inference:** Optimized for 15-20 second rendering
5. **Continuous learning:** Models improve over time with usage

## Technical Infrastructure

### Server Architecture

**Where processing happens:**

* All AI processing occurs on Looksy's cloud servers
* Not on your Shopify store
* Not on customer devices

**Benefits:**

* No performance impact on your store
* Consistent experience across devices
* Access to powerful GPU servers
* Regular model updates without customer action

### Cloud Infrastructure

**Stack:**

* GPU-accelerated servers for AI inference
* Content delivery network (CDN) for fast global access
* Auto-scaling to handle traffic spikes
* Multi-region deployment for low latency

**Uptime:** 99.9% SLA (Pro plan)

### Data Flow

```
1. Customer device → Looksy API (HTTPS)
2. Looksy API → AI processing servers (internal)
3. AI servers → Result generation
4. Result → Customer device (HTTPS)
5. Image deleted (not stored)
```

**Privacy:** Customer photos are processed and immediately discarded. Not stored.

## Performance Optimization

### Speed Optimizations

**How we achieve 15-20 second rendering:**

1. **Model quantization:** Reduced model size without accuracy loss
2. **Batch processing:** Efficient GPU utilization
3. **Caching:** Product images cached after first use
4. **Edge computing:** Processing near user for low latency
5. **Asynchronous operations:** Parallel processing where possible

### Accuracy vs. Speed Trade-Offs

**Our approach:**

* Prioritize speed without sacrificing quality
* 95%+ accuracy maintained
* Real-time rendering over perfect detail
* Good enough to help purchase decisions

<Tip>
  If perfect photorealism were required, rendering would take 30+ seconds. Looksy balances accuracy and speed for optimal e-commerce experience.
</Tip>

## Handling Edge Cases

### Challenging Scenarios

<AccordionGroup>
  <Accordion title="Low-Quality Selfies">
    **Challenge:** Dark, blurry, or angled selfies

    **How we handle it:**

    * Image enhancement algorithms
    * Brightness and contrast adjustment
    * Noise reduction
    * Pose correction (to extent possible)

    **Limitation:** Very poor selfies may still produce suboptimal results
  </Accordion>

  <Accordion title="Complex Garments">
    **Challenge:** Heavy textures, sheerness, embellishments

    **How we handle it:**

    * Texture approximation
    * Pattern preservation
    * Structural understanding

    **Limitation:** Fine details may be simplified
  </Accordion>

  <Accordion title="Unusual Poses">
    **Challenge:** Profile views, angled selfies

    **How we handle it:**

    * 3D pose estimation
    * Perspective adjustment
    * Best-effort garment placement

    **Note:** Works with any angle - the AI adapts to different poses
  </Accordion>

  <Accordion title="Poor Product Images">
    **Challenge:** Busy backgrounds, bad lighting, low resolution

    **How we handle it:**

    * Background removal algorithms
    * Garment isolation
    * Edge refinement

    **Limitation:** Cannot fix fundamentally poor images
  </Accordion>
</AccordionGroup>

## Privacy and Security

### Data Handling

**Customer images:**

* Processed in memory only
* Never stored to disk
* Deleted immediately after rendering
* No usage for training (without explicit consent)

**Product images:**

* Cached temporarily for performance
* Public images (already on your store)
* Standard retention policies

**Compliance:**

* GDPR compliant
* CCPA compliant
* SOC 2 Type II certified (Pro plan)

### Encryption

**In transit:**

* TLS 1.3 encryption
* HTTPS for all communications
* Secure API endpoints

**At rest:**

* No customer data stored
* Product images encrypted
* Database encryption

<Check>
  Looksy is privacy-first. Customer photos are never stored, shared, or used for training without explicit consent.
</Check>

## Future Technology Roadmap

### Upcoming Enhancements

**Coming Soon:**

* Multi-item try-on (outfit combinations)
* Enhanced fabric texture rendering
* AR try-on (live camera mode)
* 3D garment rotation
* Custom fit recommendations
* Virtual fitting room (full outfit try-on)
* Social try-on features

<Note>
  Roadmap subject to change based on technical feasibility and merchant feedback.
</Note>

## Technical Specifications

### API Specifications

**Request format:**

```
POST /api/v1/try-on
Content-Type: multipart/form-data

{
  "product_id": "12345",
  "selfie": <image_file>,
  "options": {
    "quality": "high",
    "fast_mode": false
  }
}
```

**Response format:**

```json theme={null}
{
  "success": true,
  "result_url": "https://cdn.looksy.ai/results/abc123.jpg",
  "processing_time_ms": 3421,
  "confidence_score": 0.94
}
```

**Rate limits:**

* Standard: 100 requests/minute
* Pro: 1,000 requests/minute

### Supported Formats

**Input images:**

* JPEG, PNG
* Minimum resolution: 800px
* Maximum file size: 5MB
* Color space: RGB

**Output images:**

* JPEG (optimized for web)
* Resolution: matches input (max 2048px)
* Compression: 85% quality

## For Developers

### Webhooks

**Available events:**

* `try_on.started`
* `try_on.completed`
* `try_on.failed`

**Webhook payload:**

```json theme={null}
{
  "event": "try_on.completed",
  "timestamp": "2026-01-23T10:30:00Z",
  "data": {
    "try_on_id": "abc123",
    "product_id": "12345",
    "result_url": "https://cdn.looksy.ai/results/abc123.jpg",
    "processing_time_ms": 3421
  }
}
```

### Custom Integrations

**Pro plan features:**

* API access for custom implementations
* Webhook integrations
* White-label rendering
* Custom branding

Contact [sales@looksy.ai](mailto:sales@looksy.ai) for details.

## Common Technical Questions

<AccordionGroup>
  <Accordion title="Does processing happen on my server?">
    No. All AI processing happens on Looksy's cloud servers. Your store only handles displaying the button and result.
  </Accordion>

  <Accordion title="What if Looksy's servers are down?">
    Button gracefully degrades. If servers are unreachable, button won't appear or will show "Try-on temporarily unavailable." Your store continues working normally.
  </Accordion>

  <Accordion title="Can I use Looksy offline?">
    No. Virtual try-on requires internet connection for processing. The AI runs on cloud servers, not locally.
  </Accordion>

  <Accordion title="How do you handle traffic spikes?">
    Auto-scaling infrastructure handles surges automatically. Tested up to 10,000 simultaneous try-ons. No manual intervention needed.
  </Accordion>
</AccordionGroup>

## Next Steps

<CardGroup cols={2}>
  <Card title="Limitations" icon="triangle-exclamation" href="/about/limitations">
    Understand what virtual try-on can and cannot do
  </Card>

  <Card title="What Makes Looksy Different" icon="star" href="/about/what-makes-looksy-different">
    See how Looksy compares to competitors
  </Card>

  <Card title="Installation Guide" icon="rocket" href="/getting-started/how-to-install">
    Get started in 10 minutes
  </Card>

  <Card title="Image Requirements" icon="image" href="/product-setup/image-requirements">
    Optimize for best AI performance
  </Card>
</CardGroup>