An image of a virtualization software in a plurality of hosts is installed or upgraded by performing the steps of: in response to a user input, generating a software specification that specifies a base image and an add-on image; retrieving metadata of the base image and metadata of the add-on image; generating a desired image of the virtualization software based on the metadata of the base image and the metadata of the add-on image; and instructing each of the hosts to install the desired image of the virtualization software or upgrade a current image of the virtualization software to the desired image of the virtualization software.