AI-103 Practice Question 78: Smart Crops Thumbnail Generation
Microsoft Azure AI Apps and Agents Developer Associate (AI-103) · Test 8 · question 8 of 10
A media company lets editors upload a single photo and needs the app to automatically generate several thumbnail variants from it at different aspect ratios (a 1:1 square tile for a mobile grid and a 16:9 banner for the article hero) while keeping the main subject, which is usually a person's face, inside the frame. Editors currently crop each variant by hand, and the team wants this automated without training or maintaining a custom model.
Which capability should the app call for each uploaded photo?
Show answer and explanation
Answer:
Smart Crops is built for exactly this: given one or more aspect ratios between 0.75 and 1.80, it returns the pixel bounding box of the region that best preserves the image's most important content, giving priority to any detected faces, without needing any training data or a custom model. The app just crops to the returned coordinates for each requested ratio.
- B is wrong because multimodal embeddings solve retrieval, finding which images in a collection look similar to a query image or text. They don't produce a crop region for a single photo; there's no "similarity" question being asked here. - C is wrong because it reintroduces the exact cost the team wants to avoid: collecting labeled training images and maintaining a custom object-detection model for a problem Smart Crops already solves as a built-in, no-training feature. - D is wrong because facial landmark detection returns the geometry of one detected face (eyes, nose, mouth outline) for tasks like alignment or verification, not an aspect-ratio-aware crop region for the whole photo, and it produces nothing usable for photos where no face is detected, while Smart Crops still returns a sensible crop by falling back to the image's general area of interest.