This study investigates the role imagery preferences of medical virtual agents (VAs) for elderly users, presenting a prompt engineering (PE) framework to enhance emotional resonance and user acceptance. Through a mixed-methods approach, we developed the S-MVA (Medical Virtual Agents for Seniors) prompts model, generated six AI-driven avatars, and evaluated their emotional impact on 100 elderly participants in China by using the PrEmo2 scale. Results reveal significant differences in emotional responses across avatar designs, with avatars featuring anthropomorphic warmth (e.g., gentle smiles, family-like traits) and culturally tailored elements (e.g., lab coats, polite communication) eliciting stronger positive emotions. This research reveals the emotional effectiveness of tailored generative design in fostering trust and engagement among older adults, providing a structured framework for generating medical VAs imagery that address their emotional needs.