In this paper, the authors present the progress and final state of CAIN-21, an extensible and metadata driven multimedia adaptation in the MPEG-21 framework. CAIN-21 facilitates the integration of pluggable multimedia adaptation tools, automatically chooses the chain of adaptations to perform and manages its execution. To drive the adaptation, it uses the description tools and implied ontology established by MPEG-21. The paper not only describes the evolution and latest version of CAIN-21, but also identifies limitations and ambiguities in the description capabilities of MPEG-21. Therefore, it proposes some extensions to the MPEG-21 description schema for removing these problems. Finally, the pros and cons of CAIN-21 with respect to other multimedia adaptation engines are discussed.