A process generates searchable content for visual media files. The process uses partial schemas, including a source schema and a keyword schema. The process uses workers, each specifying its input schemas and its output schemas. A dependency graph includes a node for each worker, with dependencies based on the input and output schemas. The graph includes initial workers that correspond to distinct media types, which use the source schema. The process constructs a source schema instance for each selected media file, and the process traverses nodes in the graph beginning with an initial worker process according to the media type. This begins execution of multiple workers, which construct additional schema instances. Some workers extract keywords from the source file and insert the terms into the keyword schema instance. The process stores the keyword schema instance and a link to the corresponding file in a database for subsequent media queries.