AI-901 ยท Vision and image generation solutions
Sending an image to a multimodal model
Including an image (URL or base64) in a message to a multimodal model, writing prompts that ask about it, and limits such as image size, detail level and token cost.
Key points
- Put a content array in the user message: one text part with your question and one image_url part.
- The image is a public URL or a data: URL holding base64.
- Optional detail (low, high, auto) trades cost for fidelity.
- Limits: 20 MB per image, 10 images per request.
Source: https://learn.microsoft.com/en-us/azure/foundry/openai/how-to/gpt-with-vision
Practice questions
In a chat message sent to a vision-capable model, which content part type carries the picture?
- A part with type attachment_file
- A part with type image_url
- A part with type vision_blob
- A part with type image_stream
Show answer
Answer: B. A part with type image_url
Images are passed as an image_url content part next to the text part. The other type names do not exist in the chat format.
Which statements about the detail setting of an image part are correct?
- low uses a reduced 512x512 view and fewer tokens
- high adds detailed segments and costs more tokens
- auto is the default and lets the model decide
- The request fails if detail is left out
- high is the cheaper setting for large images
Show answer
Answer: A. low uses a reduced 512x512 view and fewer tokens, B. high adds detailed segments and costs more tokens, C. auto is the default and lets the model decide
low is cheap and quick, high spends more tokens on fine detail, and auto is the default. Omitting detail is allowed, and high is never the cheaper option.
True or false: a multimodal chat model can only receive an image if it is reachable through a public URL.
Show answer
Answer: False
False. A base64 data URL embeds the image in the request, so no public hosting is needed.
Practice 8 questions on this topic in Bitzu
Short lessons, instant explanations and spaced review of what you get wrong. Free on iOS and Android.
