Question

You are building a lightweight vision app in Microsoft Foundry that reviews product listings. For each listing, the app must compare three uploaded photos of the same item and produce a single description noting any visual inconsistencies between the images. When constructing the prompt to send to your deployed multimodal model, what is the most appropriate approach?

Accepted Answer

Send all three images together in a single user message along with the text instruction to compare them and describe inconsistencies.

Answer

Send three separate requests, each containing one image, then manually merge the three separate responses in your app code.

Answer

Convert the three images to text using OCR first, then send only the extracted text to the model for comparison.

Answer

Send the first image with instructions and rely on the model to request the remaining two images automatically.