The correct answer is A multimodal learning model.

This scenario requires understanding visual content from a photo and then generating a textual explanation. That means the system consumes one modality as an image and produces another modality as text. This cross-modality capability is exactly what a multimodal approach provides, since it jointly handles vision and language to produce coherent natural language output based on visual input.

An image classification model is designed to assign labels or categories to an image and it does not generate rich natural language narratives or combine vision with language generation as described.

A time-series forecasting model focuses on predicting future values from sequential data over time and it does not interpret images or produce descriptive text about visual content.

A text-only unimodal model accepts and produces only text and cannot directly interpret images, so it does not match a workflow that starts with a photo and ends with a written overview.

Full AWS Practitioner Certification Question