Detect and label inappropriate user profile images on upload, keep orchestration simple, and allow adding/removing processing steps (watermarking, compression) later. What should you do?

This solution provides the most flexible and simple orchestration. Storing uploads in Cloud Storage is standard. Eventarc provides a robust way to trigger Workflows from Cloud Storage events. Workflows excel at orchestrating complex, multi-step processes like image processing, allowing easy addition or removal of steps (e.g., watermarking, compression, Vision API calls) without re-architecting. Cloud Functions are suitable for individual processing steps. The other options are less ideal: Having a backend server fetch each image is less scalable and introduces a bottleneck. Triggering a Cloud Function that calls Cloud Run services adds unnecessary complexity and latency compared to a direct Workflows orchestration. Using GKE microservices for each step is overkill for simple image processing and adds significant operational overhead compared to serverless Workflows and Cloud Functions.