I think Create ML is for people who are just getting their feet wet with machine learning. If you already know Keras, stick with that since it gives you much more control over the model and what it learns.
The image classification model in Create ML uses something called the "vision feature print" model, which seems to be a very powerful feature extractor. It's only natural that such a powerful model will overfit on a small dataset (which I assume you're using).
But Create ML (at this point) doesn't give you any tools to regulate the training, to prevent overfitting, etc. So if this "vision feature print" model is too extreme for your use case, then using Create ML does not make a lot of sense right now.