You are a machine learning engineer tasked with building a TensorFlow text-to-image generative model. Your dataset contains billions of images along with their respective captions. You aim to design a low-maintenance, automated workflow, which must: (1) Read data from a Cloud Storage bucket, (2) Collect statistics, (3) Split the dataset into training, validation, and test datasets, (4) Perform necessary data transformations, (5) Train the model using the training and validation datasets, and (6) Validate the model using the test dataset. Given these requirements, which approach should you choose?