Multi-Objective Preference Optimization: Improving Human Alignment of Generative Models