# AI Image Analysis Enrichment

AI Image Analysis properties use computer vision to identify visual characteristics from product photos, creating valuable structured data that enhances your catalog and improves product discovery.

#### Before you begin

* Map your feed's image URL field to the system property **"image"** in your feed configuration
* Create the output properties where visual attributes will be stored:
  * Use **String type** for complex data like color breakdowns with percentages
  * Use **List type** for categorical analyses (e.g., style categories) to guide the AI into pre-defined options specific to your business
* Ensure image URLs are publicly accessible (HTTPS) and show clear product photos

#### Understanding image analysis properties

Image analysis properties work differently from text-based AI enrichment. Instead of analyzing product titles or descriptions, these properties examine actual product images to identify visual characteristics like:

* **Color composition** - Primary, secondary, and accent colors with coverage percentages
* **Visual style** - Design aesthetics, patterns, and style categories
* **Material appearance** - Texture and material identification from visual cues
* **Structural features** - Shape, silhouette, and design elements

The AI analyzes each product's image individually, making this enrichment method ideal for properties that require visual context rather than text interpretation.

***

## Create your image analysis property

Navigate to **Product Data > Properties** and click **Add Property** to create a new image analysis property.

### Configure basic settings

1. Enter a descriptive **Property Name** (e.g., "Color Details" or "Visual Style")
2. A **Property Key** will automatically pre-fill the field based on the name
   1. Property keys should be unique, short, and identifiable (e.g., "color\_details" or "visual\_style")
3. Set the **Value Type** to **Image Analysis (AI)**
4. Choose your **Target Level**:
   * **Product** - For attributes consistent across all variants
   * **Variant** - For attributes that differ by variant (e.g., if your variants are defined by different colors, select this so the enricher can analyze each image variant individually)
5. Configure additional settings:
   * Turn on **Lock being edited in the app** if you want to prevent manual overrides of AI analysis
   * Turn on **Mark as internal** if you want to keep visual analysis data internal

***

### Configure AI enrichment settings

{% stepper %}
{% step %}
**Enable automated enrichment**

Turn on the **Enable automated background enrichment** toggle to activate image analysis for all eligible products (defined int the property filter). Once enabled, Crobox will automatically process eligible product images during feed imports or re-indexing.

{% hint style="info" %}
Image analysis runs automatically in the background after enabling. You'll see a progress indicator showing "Property is ready for being processed in the background."
{% endhint %}
{% endstep %}

{% step %}
**Write your visual analysis prompt**

Create a clear, specific prompt that defines what visual characteristics the AI should identify. Your prompt should include:

**Essential components:**

* **Property definition** - What visual attribute to identify
* **Output format** - How to structure the results (e.g., color breakdown, style category)
* **Visual markers** - Specific visual cues the AI should look for
* **Coverage guidelines** - Percentage thresholds or prominence rules
* **Constraints** - What to include or exclude from analysis

**Example prompt for color analysis:**

```
Analyze the product image and identify visible colors with their coverage percentages.

Format: "primary_color:percentage|secondary_color:percentage|accent_color:percentage"

Instructions:
- Identify all colors visible on the main product body, accents, and details
- Calculate approximate percentage of total visible surface area for each color
- List colors in descending order by coverage
- Include only colors with ≥5% coverage
- Total percentages must sum to 100%

Use consistent color names: black, grey, navy, blue, green, olive, brown, tan, beige, orange, red, yellow, white, purple, multi-color

Coverage patterns:
- Primary: Usually 40-80%
- Secondary: Usually 15-40%
- Accents: Usually 5-20%

Example: "navy:80|white:15|gold:5"
```

**Example prompt for style identification:**

```
Classify the product's visual style based on design elements and aesthetic.

Categories: Modern, Classic, Vintage, Minimalist, Bold, Athletic, Elegant, Rustic

Consider:
- Overall design aesthetic
- Color palette complexity
- Pattern presence
- Structural elements
- Visual balance

Return only the single most appropriate category.
```

{% hint style="warning" %}
Image analysis prompts require more detail than text-based prompts. Be specific about visual markers, coverage percentages, and formatting requirements to ensure consistent results.
{% endhint %}
{% endstep %}

{% step %}
**Define inputs and outputs**

**Inputs:**

* Select **Image** as your input source
* The system automatically uses the product's primary image URL
* Ensure your product feed includes valid, accessible image URLs

**Outputs:**

* Select the property where visual analysis results will be stored
* For complex analyses (like color breakdowns), use String type properties
* For categorical analyses (like style identification), use List type properties to guide results into pre-defined categories specific to your business
  {% endstep %}

{% step %}
**Test with preview**

Validate your image analysis setup before enabling enrichment across your catalog.

1. Click **Fetch Preview Values** to analyze sample products
2. Review the preview table showing product images, titles, and AI-generated results
3. Compare results against what you see in the product images to verify accuracy

Use **Add Filter** to test on specific product subsets and identify how well the analysis works across different product types.

{% hint style="warning" %}
Be specific about visual markers, coverage percentages, and formatting requirements in your prompt to ensure consistent results.
{% endhint %}
{% endstep %}

{% step %}
**Save and monitor enrichment**

1. Click **Save** to store your configuration
2. Enrichment begins automatically in the background
3. Monitor the status indicator: **Changed** > **Active** > **Completed**

Image analysis processes approximately 100-120 products per hour, so larger catalogs may take several hours to complete.
{% endstep %}

{% step %}
**Review and refine results**

After enrichment completes, check quality by navigating to **Product Data > Catalog** and reviewing your image analysis property alongside product images. If needed, refine your prompt based on any edge cases discovered, then click **Remove enriched data** and save to re-run enrichment with improved instructions.
{% endstep %}
{% endstepper %}

***

## Best Practices

* **Prompt clarity:** Be visually specific about what to look for, not what to read. Define clear boundaries with percentage thresholds and categorical options.
* **Image quality:** Use clear, well-lit product photos with neutral backgrounds. Avoid lifestyle images, multiple products, or low-resolution photos.
* **Property organization:** Use descriptive names for properties and create separate properties for each distinct visual attribute. Use List type for pre-defined categories specific to your business.

### Prompt examples

<details>

<summary>Color palette extraction</summary>

Extract colors and coverage percentages for accurate color-based filtering and discovery.

**Prompt approach:**

```
Analyze the product and return: "primary:percentage|secondary:percentage|accent:percentage"
Include only colors with ≥5% coverage; total must sum to 100%.
```

</details>

<details>

<summary>Style classification</summary>

Categorize products by visual design aesthetic for style-based recommendations and curation.

**Prompt approach:**

```
Classify from: Modern, Classic, Vintage, Minimalist, Bold, Athletic, Elegant
Return the single best match based on design elements.
```

</details>

<details>

<summary>Pattern recognition</summary>

Detect visual patterns for pattern-based filtering and trend-based merchandising.

**Prompt approach:**

```
Identify dominant pattern: Solid, Striped, Plaid, Floral, Abstract, Geometric, Camo, Animal Print
Return "Solid" if no pattern is visible.
```

</details>

***

## Troubleshooting

{% tabs %}
{% tab title="Image analysis not processing" %}

* Verify image URLs are publicly accessible (HTTPS) and in supported formats (JPG, PNG, WebP)
* Confirm "Enable automated background enrichment" is toggled on to ensure new products receive image analysis enrichment
* Check that input is set to "Image" and output properties are configured
  {% endtab %}

{% tab title="Inconsistent or inaccurate results" %}

* Refine your prompt with more specific visual markers and edge case examples
* Verify product images are clear and well-lit
* Test with a smaller product subset first to validate approach
  {% endtab %}
  {% endtabs %}

***

## FAQs

<details>

<summary><strong>How long does image analysis take?</strong></summary>

Image analysis processes approximately 100-120 products per hour. A 1,000-product catalog typically takes 8-10 hours.

</details>

<details>

<summary><strong>What image formats are supported?</strong></summary>

JPG, PNG, and WebP formats via publicly accessible HTTPS URLs.

</details>

<details>

<summary><strong>Can I combine image analysis with text-based enrichment?</strong></summary>

We recommend ensuring your output properties are unique to either the image analysis or text-based enrichment properties to ensure you can QA enrichment values with ease. Use image analysis for visual attributes and text-based enrichment for descriptive attributes.

</details>

<details>

<summary><strong>Can I manually edit AI-generated values?</strong></summary>

Yes, unless "Lock being edited in the app" is enabled. Manual edits may be overwritten if you re-run enrichment.

</details>

<details>

<summary><strong>How do I re-process products after updating my prompt?</strong></summary>

Click "Remove enriched data" to clear results, update your prompt, and save. Enrichment restarts automatically.

</details>

<details>

<summary><strong>What happens if products don't have images?</strong></summary>

Products without valid image URLs will not be accurately analyzed during enrichment.

</details>

***

#### What's next

* Use visual attributes in Product Finder experiences for enhanced filtering within experiences and share product highlights with users in personalized Campaigns
* Complete with [AI Enricher Properties](https://docs.crobox.com/how-to-guides/product-data/ai-enricher-properties) to enrich and create comprehensive product data for your experiences
* Track enrichment performance in property and refine prompts based on results