Moderation API

The moderation endpoint allows you to analyze text and optionally images for potentially harmful content using AI moderation models through a single, unified API. This endpoint supports checking content against various categories such as sexual content, hate speech, harassment, self-harm, and violence.

Endpoint

POST https://api.llm.vin/v1/moderations

Request Format

{
  "model": "moderation-1",
  "input": "Text to analyze for moderation",
  "input_images": ["data:image/jpeg;base64,/9j/4AAQSkZJRgABAQ..."]
}

Required Parameters

Optional Parameters

Response Format

{
  "id": "modr-1716151540",
  "model": "moderation-1",
  "results": [
    {
      "flagged": true,
      "categories": {
        "sexual": false,
        "hate": true,
        "harassment": false,
        "self-harm": false,
        "sexual/minors": false,
        "hate/threatening": false,
        "violence/graphic": false,
        "self-harm/intent": false,
        "self-harm/instructions": false,
        "harassment/threatening": false,
        "violence": false
      },
      "category_scores": {
        "sexual": 0.1,
        "hate": 0.9,
        "harassment": 0.2,
        "self-harm": 0.05,
        "sexual/minors": 0.01,
        "hate/threatening": 0.3,
        "violence/graphic": 0.1,
        "self-harm/intent": 0.02,
        "self-harm/instructions": 0.01,
        "harassment/threatening": 0.15,
        "violence": 0.2
      }
    }
  ]
}

The response contains:

• id: A unique identifier for the moderation request.
• model: The ID of the model used for moderation.
• results: An array of moderation results, one for each input (text or text+image pair), containing:
  • flagged: Boolean indicating if the content violates any moderation category.
  • categories: Object with boolean values indicating whether each category was violated.
  • category_scores: Object with confidence scores (0 to 1) for each category.

Available Models

The available moderation models depend on the configuration and are listed via the /v1/models endpoint. Only models with the moderation capability set to true can be used. Use the /v1/models endpoint to discover available models and their capabilities.

Example Requests

Basic Text Moderation

curl "https://api.llm.vin/v1/moderations" \
  -H "Content-Type: application/json" \
  -d '{
    "input": "This content contains hateful language."
  }'

Batch Text Moderation

curl "https://api.llm.vin/v1/moderations" \
  -H "Content-Type: application/json" \
  -d '{
    "input": ["Text one", "Text two with inappropriate content"],
    "model": "moderation-1"
  }'

Text and Image Moderation

curl "https://api.llm.vin/v1/moderations" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moderation-1",
    "input": "Text to moderate",
    "input_images": ["data:image/jpeg;base64,/9j/4AAQSkZJRgABAQ..."]
  }'

Error Handling

The API returns standard HTTP status codes to indicate success or failure:

Error responses include a JSON object with details:

{
  "error": {
    "message": "Model 'gpt-4.1' does not support moderation",
    "type": "invalid_request_error",
    "code": "unsupported_capability"
  }
}

Notes

• The default moderation model is used if no model parameter is specified, provided one is configured and available.
• Image moderation is only supported by models with the image_input capability set to true.
• The moderation categories and their scores are determined by the underlying model and may vary based on the platform and model configuration.
• Ensure that input_images contains valid base64-encoded image data URLs when used.