We propose a complex rule-based system for generating Czech sentences out of tectogrammatical trees, as introduced in Functional Generative Description (FGD) and implemented in the Prague Dependency Treebank 2.0 (PDT 2.0). Linguistically relevant phenomena including valency, diathesis, condensation, agreement, word order, punctuation and vocalization have been studied and implemented in Perl using software tools shipped with PDT 2.0. Parallels between generation from the tectogrammatical layer in FGD and deep syntactic representation in Meaning-Text Theory are also briefly sketched.