Configure asset metadata generation

Available from version 2.2.6

Generate asset metadata with a configured AI model and task parameters. Configure a model as described in Model configuration. As with all tasks, the default model applies unless you override it for the generateAssetMetadata task by decoration.

Features

The generateAssetMetadata task provides: * A configurable action using jcrCommandAction * An observation listener that can auto-generate metadata when an asset changes * A Groovy script to bulk-generate metadata for assets in DAM

Model configuration

To use the gemini-2-5-flash model:

  1. Configure the model gemini-2-5-flash.

  2. Decorate <light-module-folder>/<light-module-name>/decorations/ai-accelerator-image-io/aiTasks/generateAssetMetadata.yaml:

    generateAssetMetadata.yaml
    modelId: gemini-2-5-flash

You can control which languages the task generates. By default, the task uses the Magnolia default language. To add or change languages, decorate the input as shown.

The example adds English (en), German (de), Simplified Chinese (zh-CN), and Swiss Standard German (de-CH).

generateAssetMetadata.yaml
input:
  locales:
    defaultValue:
      english: en
      german: de
      chinese: zh-CN
      swiss: de-CH

Task parameters

Set defaults by decorating the input section of aiTasks/generateAssetMetadata.yaml. You can also pass parameters when invoking the action by: * Setting the action definition’s params.taskParameters * Setting the Groovy script’s taskParameters

Parameter Description

path

Required Path to the DAM asset used to generate metadata. Use a Magnolia JCR DAM asset path.

fieldName

Required; default: description Target field to store the generated text. Allowed values: description, caption, title, subject.

prompt

Required; default: Generate a caption for the image that will be great for SEO and placed in the alt attribute of the image tag. Prompt sent to the model.

locales

Optional; default: Magnolia default language Languages to generate. Provide a map of label-to-locale code, for example english: en.

overwrite

Optional; default: false Overwrite the target field if it already has a value.

maxWords

Optional; default: 0 (no limit) Maximum number of words. 0 means unlimited.

maxLength

Optional; default: 0 (no limit) Maximum number of characters. 0 means unlimited.

maxWidth

Optional; default: 256 Maximum pixel width of the image sent to the model to reduce token usage.

Generate metadata by action

See Model configuration for model setup.

Ensure your module.yaml declares the dependency:

dependencies:
  dam-app-jcr:
    version: 4.0.0/*

The decoration below: * Adds two actions to the DAM browser subApp to write caption or description for a single asset and overwrite existing content * Enables i18n for the caption and description fields so you can view generated values in multiple languages

<light-module-folder>/<light-module-name>/decorations/dam-assets-app/apps/dam.yaml
subApps:
  jcrBrowser:
    actions:
      generateImageCaption:
        label: Generate Image Caption
        icon: icon-import
        $type: jcrCommandAction
        command: execute-task
        catalog: ai-accelerator
        params:
          taskId: ai-accelerator-image-io:generateAssetMetadata
          taskParameters:
            fieldName: caption
            overwrite: true
            maxLength: 100
            locales:
              english: en
              german: de
        availability:
          writePermissionRequired: true
          nodeTypes:
            asset: mgnl:asset
          rules:
            notDeleted:
              $type: jcrIsDeletedRule
              negate: true
      generateImageDescription:
        label: Generate Image Description
        icon: icon-import
        $type: jcrCommandAction
        command: execute-task
        catalog: ai-accelerator
        params:
          taskId: ai-accelerator-image-io:generateAssetMetadata
          taskParameters:
            fieldName: description
            overwrite: true
            maxWords: 150
            locales:
              english: en
              german: de
        availability:
          writePermissionRequired: true
          nodeTypes:
            asset: mgnl:asset
          rules:
            notDeleted:
              $type: jcrIsDeletedRule
              negate: true
    # Add the defined actions to the action bar.
    actionbar:
      sections:
        asset:
          groups:
            generate:
              items:
                - name: generateImageDescription
                - name: generateImageCaption
  # Enable i18n for caption and description.
  jcrDetail:
    form:
      properties:
        caption:
          i18n: true
        description:
          i18n: true

Generate metadata by observation

The module provides an observation listener that auto-generates metadata when an asset is added in the Assets app. The listener is disabled by default.

As of version 2.2.8, the updateMetadataOnChange flag is deprecated. Use observationMetadataGeneration instead.

When enabled, the ai-accelerator-image-io:generateAssetMetadata task first uses values defined under its input properties. If none are set, it falls back to the defaultValue entries. You can view these in the Definitions app.

Default configuration (provided by the module)
observationMetadataGeneration:
  enabled: false
  observationDelay: 1000
  fields:
    caption:
      enabled: false
      taskId: ai-accelerator-image-io:generateAssetMetadata
      taskParameters:
        fieldName: caption
        maxLength: 100
    description:
      enabled: true
      taskId: ai-accelerator-image-io:generateAssetMetadata
      taskParameters:
        fieldName: description
        maxWords: 150

Set observationMetadataGeneration.enabled: true to enable the listener and use the defaults above. By default, caption is disabled and description is enabled. You can add fields entries for title and subject.

To configure observation, decorate <light-module-folder>/<light-module-name>/decorations/ai-accelerator-image-io/config.yaml.

Example decoration (enable observation, increase delay, generate multiple locales)
observationMetadataGeneration:
  enabled: true
  observationDelay: 5000
  fields:
    caption:
      enabled: true
      taskParameters:
        maxLength: 100
        locales: &locales
          english: en
          german: de
          chinese: zh-CN
          swiss: de-CH
    description:
      enabled: true
      taskParameters:
        maxWords: 150
        locales: *locales
Table 1. Observation configuration
Property Description

enabled

Enable or disable the observation listener. Default: false.

observationDelay

Debounce delay in milliseconds before running the task after an event. Default: 1000.

fields

Map of field configurations keyed by field name (caption, description, and optionally title, subject). See .Field properties (per entry under observationMetadataGeneration.fields).
Table 2. .Field properties (per entry under observationMetadataGeneration.fields)
Property Description

enabled

Whether this field is generated by observation. Defaults: caption → false, description → true.

taskId

Task to execute. Default: ai-accelerator-image-io:generateAssetMetadata.

taskParameters

Parameters forwarded to the task. See Task parameters. Common examples include: - fieldName: target property, such as caption or description - maxWords or maxLength: limits for the generated text - locales: map of label-to-locale codes to generate multiple languages (for example, english: en)

Generate metadata by Groovy script

The ai-accelerator-image-io module includes a Groovy script to generate description properties for all images in the DAM workspace.

Open the Groovy app at ${webapp-context-path}/.magnolia/admincentral#app:groovy:browser;/ai-accelerator/CreateDescriptionForAllAssets.

import info.magnolia.ai.task.AiTaskExecutor
import info.magnolia.objectfactory.Components
import com.google.common.collect.ImmutableMap

workspace = 'dam'
nodeType = "mgnl:asset"
folderType = "mgnl:folder"
folderPath = "/"              // Narrow the scope if needed.
excludedFolders = [ ]         // Add full folder paths to skip AI lookups.
// e.g. excludedFolders = ["/travel-demo/social-icons", "/svg-icons", "/destinations"]

def generateDescription(assetPath) {
    AiTaskExecutor taskExecutor = Components.getComponent(AiTaskExecutor.class)

    String prompt = "Create a great alt text for SEO purposes"

    Map<String, Object> taskParameters = ImmutableMap.<String, Object> builder()
            .put("prompt", prompt)
            .put("path", assetPath)
            .put("overwrite", false) // Set true to overwrite existing data.
            // .put("fieldName", "caption") // Default is "description"; allowed: "caption", "title", "subject"
            .build()

    Map<String, Object> taskContext = ImmutableMap.<String, Object> builder()
            .put("taskParameters", taskParameters)
            .build()

    try {
        Map<String, Object> execute = taskExecutor.execute("ai-accelerator-image-io:generateAssetMetadata", taskContext)
        Map<String, Object> response = ImmutableMap.<String, Object> builder().putAll(execute).put("prompt", prompt).build()

        if (response.get("error") == true) {
            println response.toString()
        } else {
            LinkedHashMap responseData = ((LinkedHashMap) response.get("data"))
            println "Retrieved Description properties for: " + assetPath
            println "\tEnglish: " + responseData.get("text_en").toString()
            println "\tGerman: " + responseData.get("text_de").toString()
        }
    } catch (Exception e) {
        println "Error generating description for: " + assetPath + " " + e.toString()
    }
}

def assessFolder(targetFolder) {
    targetNode = session.getNode(targetFolder)
    def currentPath = targetNode.getPath()
    println "Analyzing folder: " + currentPath

    // Skip excluded folders.
    if (excludedFolders.any { currentPath.startsWith(it) }) {
        println "Skipping excluded folder: " + currentPath
        return
    }

    targetNodeAssets = NodeUtil.asList(NodeUtil.getNodes(targetNode, nodeType))
    targetNodeFolders = NodeUtil.asList(NodeUtil.getNodes(targetNode, folderType))

    for (assetNode in targetNodeAssets) {
        generateDescription(assetNode.getPath())
    }
    for (folderNode in targetNodeFolders) {
        assessFolder(folderNode.getPath())
    }
}

ctx = MgnlContext.getInstance()
session = ctx.getJCRSession(workspace)
assessFolder(folderPath)
Feedback

DX Core

×

Location

This widget lets you know where you are on the docs site.

You are currently perusing through the AI Accelerator module docs.

Main doc sections

DX Core Headless PaaS Legacy Cloud Incubator modules