All Products
Search

This Product

    AI Guardrails:Image Moderation Enhanced Edition console guide

    Document Center

    AI Guardrails:Image Moderation Enhanced Edition console guide

    Last Updated:Mar 31, 2026

    Image Moderation V2.0 provides preset detection configurations built on extensive content governance experience and industry practices. Use the console to customize detection rules, manage image and vocabulary libraries, test moderation performance, and query results—without writing any API code.

    Use cases

    ScenarioWhat you can doRecommended action
    Adjust the detection scopeTurn individual detection items on or off, and set confidence score thresholds for medium-risk and high-risk levels. The adjustable scope for Baseline Check (baselineCheck_global) includes: pornographic content, suggestive content, terrorist content, prohibited content, flag content, undesirable content, and abusive content.Tag or restrict content
    Support multiple business scenariosCopy a service to create scenario-specific detection configurations. For example, copy baselineCheck_global to create baselineCheck_global_01 and apply different detection scopes to each.Separate moderation policies
    Flag specific known imagesUpload images to a custom image library and associate it with a detection service. When a submitted image matches one in the library, a risk label is returned immediately. Useful for sudden events, traffic-diversion ad images, or community-specific content like cyberbullying images.Block or remove content
    Exempt trusted imagesAssociate a trusted image library with a detection service. When a submitted image matches one in the library, the system returns nonLabel_lib and skips further risk detection. Use this for verified marketing materials, official images, or manually reviewed profile pictures.Allow without re-review
    Configure text detection in imagesSet up custom vocabularies to ignore or flag specific keywords found in image text.Tag or suppress alerts
    Test moderation onlineSubmit image URLs or upload local files to preview moderation results before deploying to production. Test up to 100 images at a time.Validate configuration
    Query detection resultsSearch recently detected images by request ID, data ID, service, or returned label.Audit and provide feedback
    View usage statisticsReview call volumes and label hit distributions to guide your moderation strategy.Monitor and refine

    Prerequisites

    Before you begin, ensure that you have:

    Review the billing information for Image Moderation 2.0 before activating. For details, see V2.0V2.0V2.0V2.0V2.0V2.0V2.0V2.0V2.0V2.0V2.0Version 2.0Version 2.0Introduction to Image Moderation Enhanced Edition 2.0 and its billing information.

    Adjust the detection scope for images

    Customize which content categories are checked and set confidence score thresholds to determine the returned risk level.

    How confidence scores map to risk levels:

    The risk level returned for an image is determined as follows:

    1. If a risk label is detected and its confidence score falls within the high-risk score range → high

    2. If a risk label is detected and its confidence score falls within the medium-risk score range → medium

    3. If a risk label is detected and its confidence score falls below the medium-risk score range → low

    4. If multiple labels with different risk levels are hit → the highest risk level is returned

    5. If no risk labels are hit → safe

    6. If a custom blocklist is hit → high

    For example, if an image hits both a medium-risk label and a high-risk label simultaneously, the system returns high.

    Steps:

    1. Log on to the Content Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation console.

    2. In the navigation pane, choose Machine Moderation V2.0 > Image Moderation > Rules.

    3. On the Rules Management tab, find the service to configure—this example uses Common Baseline Moderation (baselineCheck_global)—and click Settings in the Actions column._global_global_global

    4. On the Detection Scope page, select a detection type. This example uses Prohibited Content Detection.

      1. On the Prohibited Content Detection tab, review the default configurations in the Detection Scope Configuration section. The following figure shows four items detected by default.

    image

    1. Click Edit and change the Detection Status for each item as needed. The following figure shows the detection switch for the third item turned off.

    image

    1. Adjust the Medium-risk Score and High-risk Score thresholds to control which confidence score ranges map to each risk level.

    2. Click Save. The new configuration takes effect in about 2 to 5 minutes.

    Set different detection scopes for multiple business scenarios

    Copy an existing service to create independent configurations for different business scenarios.

    1. Log on to the Content Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation console.

    2. In the navigation pane, choose Machine Moderation V2.0 > Image Moderation > Rules.

    3. On the Rules Management tab, find the service to copy—this example uses Common Baseline Moderation (baselineCheck_global)—and click Copy in the Actions column.

    4. In the Copy Service panel, enter a Service Name and Service Description.

    image.png

    1. Click Create. The copied service (for example, baselineCheck_global_01) is available within 1 to 2 minutes.

    2. Configure the new service with the detection scope required for that scenario. Call it independently of the original service to meet different moderation requirements.

    Flag specific known images

    Configure a custom image library so that any submitted image matching one in the library is immediately flagged with a risk label.

    Each account can create up to 10 image libraries. The total number of images across all libraries cannot exceed 100,000.

    Create an image library and upload images

    Skip this section if an existing library already meets your requirements.

    1. Log on to the Content Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation console.

    2. In the navigation pane, choose Machine Moderation V2.0 > Image Moderation > Image Libraries.

    3. Click Create Image Library. In the Create Image Library dialog box, enter a library name and description, then click OK.

    4. Find the library you created and click Image Detail in the Actions column.

    创建图库

    1. Click Add Image. In the Add Image dialog box, select images to upload.

      • Upload up to 10 images at a time; each image must be 4 MB or smaller.

      • The minimum recommended resolution is 256 × 256 pixels.

      • The upload list shows up to 10 images. To upload more, clear the list and continue.

    添加图片-选择图片

    1. On the library details page, view uploaded images. Search by Image ID or Add Time, or delete images individually or in bulk.

    图库管理-上传图片

    Maintain an existing image library

    1. In the navigation pane, choose Machine Moderation V2.0 > Image Moderation > Image Libraries.

    2. Find the library to maintain. Click Edit in the Actions column to update the name and description. Click Image Detail to upload or delete images.

    创建图库

    Associate the image library with a detection service

    1. In the navigation pane, choose Machine Moderation V2.0 > Image Moderation > Rules.

    2. On the Rules Management tab, find the service to configure—this example uses Common Baseline Moderation (baselineCheck_global)—and click Settings in the Actions column.

    3. On the Detection Scope page, select a detection type. This example uses Prohibited Content Detection.

    4. In the Set Labels by Customized Libraries section of the Prohibited Content Detection tab, view the currently configured custom image library.

    image

    1. Click Edit and select the image library to associate.

    image

    1. Click Save. The configuration takes effect in about 2 to 5 minutes.

    When a submitted image matches an image in the associated library, the contraband_drug_lib label is returned.

    Exempt trusted images from risk detection

    Associate a trusted image library with a detection service so that matched images bypass risk detection entirely.

    1. Log on to the Content Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation console.

    2. In the navigation pane, choose Machine Moderation V2.0 > Image Moderation > Rules.

    3. On the Rules Management tab, find the service to configure and click Settings in the Actions column.

    4. Click the Exemption Configuration tab.

    5. View the list of custom image libraries and their exemption status. The following figure shows all exemption switches turned off.

    免审图配置前.png

    1. Click Edit and turn on the exemption switch for the required library.

    免审图配置后

    1. Click Save. The exemption configuration takes effect in about 2 to 5 minutes.

    The system compares each submitted image against the images in the exempted library for similarity. For images identified as identical, the system returns nonLabel_lib and no other risk labels.

    Configure text detection in images

    Set up custom vocabularies to ignore or flag specific keywords found in image text.

    Create a custom vocabulary

    Skip this section if an existing vocabulary already meets your requirements.

    A single account can have up to 20 vocabularies with a total of 100,000 keywords. A single keyword cannot exceed 20 characters. Special characters are not supported.

    1. Log on to the Content Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation console.

    2. In the navigation pane, choose Machine Moderation V2.0 > Text Moderation > Library Management.

    3. On the Keyword Library Management tab, click Create Library.

    4. In the Create Library panel, enter the vocabulary information and click Create Library. You can also create a vocabulary without adding keywords and add them later as needed. If the vocabulary fails to be created, an error message is displayed. Retry based on the message.

    Configure an ignored vocabulary

    Add keywords that image text detection should skip, so images containing those keywords are not flagged as non-compliant.

    1. In the navigation pane, choose Machine Moderation V2.0 > Image Moderation > Rules.

    2. On the Rules Management tab, find the service to configure—this example uses Common Baseline Moderation (baselineCheck)—and click Settings in the Actions column.

    3. On the Ignoring vocabulary configuration tab, view the list of custom vocabularies and their status. The following figure shows all switches turned off.

    image

    1. Click Edit and turn on the switch for the vocabulary to ignore.

    image

    1. Click Save. The configuration takes effect in about 2 to 5 minutes.

    The system excludes the vocabulary keywords before performing risk detection. For example, if the text in an image is "Here is a small cat" and the vocabulary contains "is" and "a", risk detection runs only on "Here small cat".

    Configure a flagged vocabulary

    Add keywords that trigger a risk label when found in image text.

    1. In the navigation pane, choose Machine Moderation V2.0 > Image Moderation > Rules.

    2. On the Rules Management tab, find the service to configure and click Settings in the Actions column.

    3. On the Detection Scope page, select a detection type. This example uses Prohibited Content Detection.

    4. In the Set Labels by Customized Libraries section of the Prohibited Content Detection tab, view the configured custom vocabulary.

      The Set Labels by Customized Libraries section supports all labels ending with tii. The tii suffix indicates that a risk was detected in the text of an image.
    image

    1. Click Edit and select the custom vocabulary to associate.

    image

    1. Click Save. The configuration takes effect in about 2 to 5 minutes.

    When the text in a submitted image matches a keyword in the associated vocabulary, the contraband_drug_tii_lib label is returned.

    Test image moderation online

    Submit images directly in the console to preview moderation results before updating your production configuration.

    1. Log on to the Content Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation console.

    2. In the navigation pane, choose Machine Moderation V2.0 > Online Test.

    3. On the Online Test page, click the Image tab.

    4. From the Service drop-down list, select the service to test.

      For accurate results, adjust the service rules on the Rules page before testing.

    5. Optionally, enter a DataId and Auxiliary Information. For parameter descriptions, see the Image moderation API documentation.

    6. Provide images by entering an Image URL or by uploading a local image. Submit up to 100 images at a time.

    7. Click Test. The moderation results appear in the result area.

    Query detection results

    Search recently detected images to analyze individual moderation decisions or submit feedback on results.

    1. Log on to the Content Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation console.

    2. In the navigation pane, choose Machine Moderation V2.0 > Image Moderation > Detection Results.

    3. On the Detection Results page, enter query conditions to filter results. Supported conditions: request ID, data ID, service, and returned label.

      • To search by label, enter one or more labels separated by commas (,) in the Returned Label field. For example, set Returned Label to !=nonLabel to return all records with a hit label.

      Results are displayed in reverse chronological order. Up to 50,000 entries are shown, covering the last 30 days. Store your API call data or logs to support analysis over longer periods.
    image.png

    1. Click an image or click Details in the Actions column to view full detection information.

    image..png

    1. To report an incorrect result, select False Positive or Missed Violation from the Feedback drop-down list in the Actions column.

    image.png

    View usage statistics

    Review detection volumes and label distributions to understand moderation patterns and refine your configuration.

    1. Log on to the Content Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation console.

    2. In the navigation pane, choose Machine Moderation V2.0 > Image Moderation > Usage Statistics.V2.0

    3. On the Usage Statistics page, select a time range. Statistical data is stored for one year; queries can span up to two months.

      • Query usage: View daily or monthly call volumes.

    导出用量

    • Export usage: Click the Download icon to export daily or monthly data as an Excel file. The exported report includes only data with call volumes and contains the following fields:

    FieldDescriptionUnit
    Account UIDThe UID of the account that exported the data—
    ServiceThe called detection service—
    UsageThe total number of callsCount
    DateThe date the statistics were collectedDay/Month

    • View service hit details: After usage statistics are generated, label hit details for each called service are displayed as two charts:

      • Column chart of daily call volumes: Shows the number of daily requests that hit risk labels versus those that did not.

      • Treemap chart of label proportions: Shows overall label hit distribution in descending order of proportion. Labels sharing the same prefix have the same background color.

    image.png