In this paper, we discuss the study and implementation of two alternative artificial intelligence methodologies in declarative scene synthesis. In order to achieve this, we employ two approaches towards the acquisition and exploitation of the desired qualitative aspects of the synthesis on behalf of the designer. Our work concerns the adaptation and application of a machine learning technique, as well as of an evolutionary search method, in the current context. The former refers to the gradual construction of a preference model, comprising an incrementally learning mechanism based on actual designer solution evaluation during regular system use. The latter approach models qualitative aspects through a multi-objective genetic algorithm variation based on the weighted sum method. The algorithm is applied during the generation and understanding phases of Declarative Modeling. This work provides evidence of performance improvement of declarative scene synthesis at the expense of additional designer feedback.