AI-103 Practice Question 7: Mask Based Image Editing
Microsoft Azure AI Apps and Agents Developer Associate (AI-103) · Test 1 · question 7 of 10
A furniture retailer has 4,000 existing studio photographs of sofas. Marketing wants to swap the plain grey backdrop in each photo for a styled living-room scene while leaving the sofa itself pixel-for-pixel untouched, since the product must remain a faithful representation of what ships.
Which approach in Microsoft Foundry meets that requirement?
Show answer and explanation
Answer:
The Image Edit API modifies an existing image rather than producing a new one. You pass the source image plus a mask, which defines the area the model may edit: the mask must be a PNG of the same dimensions as the input image, with fully transparent pixels (alpha of zero) marking the region to change. Masking the backdrop and leaving the sofa opaque is exactly the inpainting workflow this scenario calls for.
- A is wrong because the generations endpoint synthesizes a brand-new image from text alone. However carefully the prompt describes the sofa, the result is a model's rendering of a sofa, not the photographed product. - B is wrong for the same reason, with an extra lossy step: a caption cannot carry stitch detail, fabric weave, or exact colour, so the regenerated sofa would not match what ships. - D is wrong because Content Understanding extracts structured metadata and fields from media. It is an analysis capability, not an image editor, and it produces no edited image.