The field of text-to-image generation is moving towards personalized and explainable models that can align with human perception and preferences. Researchers are developing new methods for evaluating and optimizing generated images to better match individual user tastes. This includes the introduction of datasets and benchmarks that capture diverse user preferences and the development of models that can dynamically generate user-conditioned evaluation dimensions. Noteworthy papers in this area include IE-Critic-R1, which introduces a comprehensive and explainable quality assessment metric for text-driven image editing, and MagicWand, a universal generation and evaluation agent that enhances prompts based on user preferences. Additionally, PIGReward, a personalized reward model, and RubricRL, a simple and general framework for rubric-based reward design, demonstrate promising approaches to personalized text-to-image generation and evaluation.