Multimodal embedding

Parameter

Type

Required

Description

Example

Content-Type

String

Yes

Content-Type: application/json

application/json

Authorization

String

Yes

The API key for authentication.

Bearer OS-d1**2a

Parameter

Type

Required

Description

Example

input

List[ContentObject]

Yes

The content to embed. For qwen3-vl-embedding, each request supports up to 20 content elements, including up to 10 images and 1 video. The other listed models support up to 32 input items per request. When options.enable_fusion is false, a separate vector is generated for each ContentObject. When it is true, all content in input is fused into a single vector. enable_fusion applies only to qwen3-vl-embedding.

[
  {
    "text":"Science and technology are the primary productive forces"
  },
  {
    "image":"http://***/a.jpg"
  }
]

options

OptionObject

No

Model settings, including image resizing and vector output settings. For the models to which each field applies, see OptionObject.

{
  "enable_fusion": false,
  "dimension": 1024
}

Parameter

Type

Required

Description

Example

text

String

No

Text to convert into an embedding.

{
  "text":"Text input"
}

image

String

No

The image information, specified as a URL or Base64-encoded data.

• URLs must be publicly accessible.

• For Base64 input, pass data to the image parameter in the format data:image/{format};base64,{base64_image}, where:

  image/{format}: The local image format. Set this value to the actual image format. For example, for a JPG image, use image/jpeg.

  base64_image: The Base64-encoded image data.

{
  "image":"http://xxxxx/a.jpg"
}

or

{
  "image":"data:image/jpeg;base64,/9j/4AAQSkZJRgABAQAAAQABAAD/2wCEAAoHCB..."
}

video

String

No

The URL of a video file. Only qwen3-vl-embedding supports this parameter. Setting video for other models causes an error.

{
  "video": "https://example.com/video.mp4"
}

Parameter

Type

Required

Description

Example

resize_method

String

No

The image resizing method:

• default: Uses the default resizing behavior built into the model.

• none: Does not resize the image.

• fixed_size: Resizes the image to the width and height specified in resize_options.

• proportional_scaling: Preserves the aspect ratio and resizes the image to the pixel range specified in resize_options.

fixed_size

resize_options

Object

No

The resizing parameters used when resize_method is fixed_size or proportional_scaling. For fixed_size, you must specify width and height in pixels. For proportional_scaling, use min_pixels and max_pixels to specify the minimum and maximum pixel counts after resizing.

fixed_size:

{
  "width": 224,
  "height": 224
}

proportional_scaling:

{
  "min_pixels": 409600,
  "max_pixels": 1638400
}

enable_fusion

Boolean

No

Specifies whether to generate a fused vector. This parameter applies only to qwen3-vl-embedding and defaults to false. If false, a separate vector is generated for each input item. If true, all content in input is fused into a single vector. Supported combinations include text and images, text and video, multiple images and text, and images, video, and text.

true

instruct

String

No

A custom task instruction that guides the model in understanding the query intent. This parameter applies only to qwen3-vl-embedding. English instructions are recommended.

Represent the input for cross-modal retrieval.

dimension

Int

No

The output vector dimension. qwen3-vl-embedding supports 2560, 2048, 1536, 1024, 768, 512, and 256. The default is 2560.

1024