CCDV-F Domain 4: Vision & Multi-Modal Integration

Passing images to the Messages API, base64 payload structure, supported media formats, token cost estimation, and document extraction best practices.

1. Vision Content Block Schema

Claude 3 and 3.5 models natively accept images within the content array of a message object:

{
  "role": "user",
  "content": [
    {
      "type": "image",
      "source": {
        "type": "base64",
        "media_type": "image/jpeg",
        "data": "/9j/4AAQSkZJRgABAQEASABIAAD..."
      }
    },
    {
      "type": "text",
      "text": "Transcribe the financial table from this receipt image into CSV format."
    }
  ]
}

2. Supported Image Formats & Size Limits

• Supported Media Types: image/jpeg, image/png, image/gif, and image/webp.
• Size Limits: Up to 5 MB per image for base64 encoded strings (or up to 100 images per request in batch arrays).
• Resolution: Optimal resolution is between 1500px and 2000px on the longest edge. Downscale massive multi-megapixel images before base64 encoding to conserve tokens and reduce upload latency.

3. Vision Token Calculation