You have a Microsoft Foundry project that contains a vision-enabled model deployment. You are developing an application that sends images to the…

Question breakdown

Tests

Send image input to a vision-enabled model in Microsoft Foundry

Difficulty

Easy

Key idea: Vision-enabled models need the actual image, supplied either as a URL the service can fetch or as base64-encoded bytes embedded in the request.

How to reach the answer

1. The model must 'analyze the images', so it needs real pixel data, not a summary of the image.
2. List the ways a chat or responses request can carry an image: by reference or inline.
3. By reference means a URL the service can reach and download.
4. Inline means the bytes are base64-encoded, usually as a data URL, inside the JSON payload.
5. Text or JSON that only describes the image contains no pixels, so rule out A and B.

Each option

• AIncorrect. A JSON description holds no image data, so the model would only be reading text about the picture.
• BIncorrect. A UTF-8 caption is plain text; the model cannot do visual analysis without the image itself.
• CCorrect. A publicly reachable image URL lets the service download the image and analyze it.
• DCorrect. Base64-encoded image data embeds the actual bytes directly in the request payload.

Common trap: Because the request body is JSON, it is tempting to think a JSON or text description of the image is acceptable, but the model needs the pixels.

In practice: Apps typically pass blob or CDN URLs for images that are already hosted. They base64-encode local files, user uploads, and private images that the service cannot reach over the internet.

Remember: Give vision models images as a reachable URL or as base64-encoded data, never as a text description.

Official documentation: Use vision-enabled chat models

Drafted with AI assistance and checked against the answer key.