Among existing tools for laying out large collections of visual objects, some perform automatic layouts, possibly following some rules prespecified by the user, e.g., graph layout tools, while others let users specify layouts manually, e.g., CAD design tools. Most of them can only deal with specific types of visualizations, e.g., graphs, and some of them allow users to view visual objects at various levels of detail, e.g., tree-structure visualization tools. In this paper, we develop techniques that strike a balance between user specification and automatic generation of layouts, work at multiple granularities, and are generally applicable. In particular, we introduce a general framework and layout algorithm that (a) deals with arbitrary types of visual objects, (b) allows objects to be viewed in any one of several different visual representations (at different levels of detail), and (c) uses a small number of user-specified layouts to guide heuristic decisions for automatically deriving many other layouts in a manner that attempts to be consistent with the user's preferences. The algorithm has been implemented within the OPOSSUM database schema manager and has been rather effective in capturing the intuition of scientists from several disciplines who have used it to design their database and experiment schemas.