Visual Understanding

Models with vision support (such as gpt-5.6, fable-5, omni-flash) can directly understand images in a conversation: write content as a multimodal array, mixing text and image_url.

Request Example

curl https://api.4allapi.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AIPROXY_KEY" \
  -d '{
    "model": "omni-flash",
    "messages": [{
      "role": "user",
      "content": [
        {"type": "text", "text": "What is the amount and invoice date of this invoice?"},
        {"type": "image_url", "image_url": {"url": "https://example.com/invoice.jpg"}}
      ]
    }]
  }'

Images can also be embedded using a base64 Data URL (suitable for intranet images):

{"type": "image_url", "image_url": {"url": "data:image/jpeg;base64,/9j/4AAQ..."}}

Typical Scenarios

• Structured information extraction from receipts / tables / screenshots (use response_format to force JSON output);
• Tagging and compliance review for e-commerce product images;
• Converting UI screenshots to code, interpreting chart data;
• Comparing multiple images: place multiple image_url entries sequentially in the content array.

Notes

• Images are billed by token equivalence based on resolution; for very large images, it is recommended to compress them to within 2048px first;
• The URL method requires the image to be publicly accessible; for intranet or temporary images, use base64;
• For video understanding, use a model that supports video input and refer to its model documentation.