You are constructing an actual model by selecting some layers from the given vgg model. Then you invoke that model with two different inputs: the content image and the style image.