Skip to main content
Doubao Embedding Vision is ByteDance’s multimodal embedding model, available through Upmore via the standard embeddings API (/v1/embeddings). It maps text and images into a single 2048-dimension vector space — ideal for image-text retrieval and semantic search.

​
Key capabilities

  • Multimodal input — Embed text, an image, or a combined text+image input into one vector
  • Unified vector space — Text and image vectors are directly comparable
  • 2048 dimensions — Fixed output dimension
  • Split billing — Text input and vision input are billed at separate rates; usage is reported per type

​
Quick example

curl https://api.upmore.net/v1/embeddings \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-embedding-vision",
    "input": [
      {"type": "text", "text": "A ginger cat napping on a piano"},
      {"type": "image_url", "image_url": {"url": "https://example.com/cat.jpg"}}
    ]
  }'

import requests

response = requests.post(
    "https://api.upmore.net/v1/embeddings",
    headers={
        "Authorization": "Bearer YOUR_API_KEY",
        "Content-Type": "application/json",
    },
    json={
        "model": "doubao-embedding-vision",
        "input": [
            {"type": "text", "text": "A ginger cat napping on a piano"},
            {"type": "image_url", "image_url": {"url": "https://example.com/cat.jpg"}},
        ],
    },
)

result = response.json()
embedding = result["data"]["embedding"]  # 2048 floats
print(len(embedding), result["usage"])

const response = await fetch('https://api.upmore.net/v1/embeddings', {
  method: 'POST',
  headers: {
    Authorization: `Bearer ${process.env.UPMORE_API_KEY}`,
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    model: 'doubao-embedding-vision',
    input: [
      { type: 'text', text: 'A ginger cat napping on a piano' },
      { type: 'image_url', image_url: { url: 'https://example.com/cat.jpg' } },
    ],
  }),
})

const result = await response.json()
console.log(result.data.embedding.length, result.usage)
Unlike standard embedding models, input is an array of content objects (Ark multimodal format), not a plain string. Text-only input also uses this format: [{"type": "text", "text": "..."}]. One request returns one fused embedding for the whole input.

​
Parameters

ParameterTypeRequiredDescription
modelstringYesMust be doubao-embedding-vision
inputarrayYesContent objects: {type: "text", text} and/or {type: "image_url", image_url: {url}}
input[].image_url.urlstringImage itemsImage URL or Base64 data URI

​
Response

FieldDescription
data.embedding2048-dimension float vector for the whole input
usage.prompt_tokens_details.text_tokensText tokens — billed at the text rate
usage.prompt_tokens_details.image_tokensImage tokens — billed at the vision rate (higher)

API Reference

View the interactive API playground.