The requirement explicitly states that the solution must not involve training a machine learning model, which immediately eliminates options that require custom ML development. Amazon Rekognition is a fully managed computer vision service that can analyze images and detect inappropriate or unwanted content using pretrained models provided by AWS.

Option B correctly uses Amazon Rekognition’s image moderation capabilities to analyze uploaded photos and identify unsafe or unwanted content categories. By invoking Rekognition from an AWS Lambda function, the solution remains serverless, automatically scalable, and operationally efficient. The Lambda function URL provides a secure HTTPS endpoint that the web application can call synchronously during uploads.

Option A violates the requirement because SageMaker Autopilot is designed to build and train ML models. Option C is incorrect because Amazon Comprehend is a natural language processing service for text analysis, not image moderation. CloudFront Functions also do not support calling external AWS services like Rekognition. Option D uses Rekognition Video, which is intended for analyzing video streams, not static images.

Using Lambda and Rekognition together ensures minimal operational overhead, no ML training effort, and immediate enforcement of content moderation policies. The architecture also allows easy extension, such as logging moderation results or blocking uploads based on confidence thresholds.

Therefore, B is the correct solution because it meets the content moderation requirement using AWS-managed ML services without requiring model training.