AI Guardrails:LLM image moderation

Step 1: Activate the service

Go to the service activation page and activate Image Moderation 2.0. After activation, billing defaults to pay-as-you-go — you are only charged for API calls that return HTTP 200.

Step 2: Grant permissions to a RAM user

Create an AccessKey pair for your Alibaba Cloud account or a RAM user. The RAM user must have the AliyunYundunGreenWebFullAccess policy to call Content Moderation APIs.

1. Log on to the RAM console using your Alibaba Cloud account or as a RAM administrator.

2. Create a RAM user. For details, see Create a RAM user.

3. Attach the AliyunYundunGreenWebFullAccess policy to the RAM user. For details, see Manage permissions for RAM users.

Step 3: Install the SDK

Follow the Image Moderation SDK and integration guide to install the SDK and configure your endpoint.

The service is available in three regions:

Step 4: Configure detection rules (optional)

In the Content Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation console, adjust detection rules: enable or disable detection categories, configure a custom image library, query detection records, and review usage data. For details, see Console guide.

API overview

• Operation: ImageModeration

• Service code: postImageCheckByVL_globalglobal_global

• QPS limit: 50 calls/second per user. Exceeding this limit throttles requests. Contact your business manager for a quota increase.

• Billing: Charged per successful request (HTTP 200). USD 1.20 per 1,000 calls, settled daily.

• Image Moderation for Large and Small Model Integration:

  Combines large and expert models to detect a wide range of non-compliant content in images, such as pornography, suggestive material, politically sensitive content, violence, terrorism, prohibited items, religious content, spam, and other undesirable content. (Note: All large model inference is processed in the Singapore region.) For details on the detection categories, see Rules.

• postImageCheckByVL_global: Image Moderation for Large and Small Model Integration

Debug the API

Test the API before integrating using Alibaba Cloud OpenAPI. The tool generates sample code and SDK dependency information.

Image requirements

Submit an image

Image Moderation 2.0 supports three ways to submit an image. Choose one per request:

Request parameters

The request body is a JSON object. For required common request parameters, see the Integration guide.

Top-level parameters

ServiceParameters fields

Request example

{
    "Service": "postImageCheckByVL_global",
    "ServiceParameters": {
        "imageUrl": "https://img.alicdn.com/tfs/TB1U4r9AeH2gK0jSZJnXXaT1FXa-2880-480.png",
        "dataId": "img0307****"
    }
}

Response parameters

Top-level response fields

Data fields

Result fields

Ext fields

CustomImage fields

Response example

{
    "RequestId": "70ED13B0-BC22-576D-9CCF-1CC12FEAC477",
    "Code": 200,
    "Msg": "OK",
    "Data": {
        "RiskLevel": "high",
        "DataId": "img0307****",
        "Result": [
            {
                "Label": "violent_explosion",
                "Confidence": 92.40,
                "Description": "Fireworks content",
                "RiskLevel": "high"
            },
            {
                "Label": "violent_burning",
                "Confidence": 67.15,
                "Description": "Burning scenes",
                "RiskLevel": "medium"
            }
        ],
        "Ext": {}
    }
}

Request and response examples are formatted for readability. Actual API responses do not include line breaks or indentation.

Understanding risk levels and handling results

Each label has an individual RiskLevel and Confidence score. The RiskLevel in the Data object reflects the highest risk across all returned labels.

Use RiskLevel to guide your moderation workflow:

Tuning confidence thresholds: The default risk score thresholds determine when a label is assigned high, medium, or low. Lowering a threshold increases recall (fewer missed violations) but also increases false positives. Raising a threshold improves precision but may miss some violations. Adjust thresholds per label in the Content Moderation consoleContent Moderation consoleContent Moderation consoleContent Moderation console to match your platform's tolerance for false positives versus false negatives.

Store the risk labels and confidence scores returned by the system for a defined period. Use them to prioritize content for manual review, build annotation datasets, and apply tiered governance policies based on risk category.

Label reference

Labels are grouped into the categories below. The _tii suffix indicates that the label applies to text detected within the image (text-in-image), rather than the visual content itself.

Pornographic content (pornographic_*)

Sexually suggestive content (sexual_*)

Politically sensitive content (political_*)

Violence and terrorism (violent_*)

Contraband (contraband_*)

Religious content (religion_*)

Flags

Spam and promotional content (pt_*)

Inappropriate behavior (inappropriate_*)

Profanity (profanity_*)

Custom image library labels

Configure a custom image library for any risk category in the console. When a submitted image is highly similar to an image in your library, the system returns the corresponding label with a _lib suffix (for example, violent_explosion_lib). The Confidence score reflects the degree of similarity.

No-risk labels