The development of therapeutic monoclonal antibodies requires optimizing multiple fitness parameters simultaneously. Beyond binding affinity, successful candidates must exhibit robust developability, stability and controlled structural dynamics. This seminar explores strategies for multi-objective antibody design, integrating both statistics- and physics-based approaches to balance these competing fitness criteria. Furthermore, we introduce a novel conditional generative model based on a fine-tuned GNN architecture. By utilizing pLDDT as a quantifiable proxy for local flexibility, this model explicitly steers sequence generation toward specified dynamic profiles. The model is shown to actively respond to conditioning through the targeted use of rigid or flexible amino acids. Molecular dynamics simulations corroborate the successful modulation of structural flexibility, revealing that mutations predominantly localize at the CDR hinges. This approach provides a new, physics-aware dimension of control for rational therapeutic engineering.