The website simply takes your image submission, runs it through the Google Vision API, and spits back out a description of the image.

Currently, in preview, Vertex AI Vision is an extension of AutoML Vision that can train models to perform image classification and object detection.