MagicWand: A Universal Agent for Generation and Evaluation Aligned with User Preference

TL;DR

MagicWand addresses the difficulty of aligning AI-generated content with individual user preferences by introducing a universal, multi-task agent built on a three-module architecture (Planner, Executor, Summarizer) that plans, executes, and reflects on prompts and outputs. It leverages UniPrefer-100K to ground learning of user styles and UniPreferBench to benchmark cross-task alignment across five AIGC modalities, with a memory-driven Preference Learner and adaptive Threshold Regulator guiding iterative refinements. Empirical results on UniPreferBench show MagicWand achieving state-of-the-art alignment for both content generation and evaluation, outperforming baselines across T2I, I2I, T2V, I2V, and V2V tasks, while maintaining interpretability through structured evaluation reasoning. The framework enables personalized, cross-task, preference-aware content creation with practical impact for developers and end users seeking consistent, preference-aligned AIGC experiences.

Abstract

Recent advances in AIGC (Artificial Intelligence Generated Content) models have enabled significant progress in image and video generation. However, users still struggle to obtain content that aligns with their preferences due to the difficulty of crafting detailed prompts and the lack of mechanisms to retain their preferences. To address these challenges, we construct \textbf{UniPrefer-100K}, a large-scale dataset comprising images, videos, and associated text that describes the styles users tend to prefer. Based on UniPrefer-100K, we propose \textbf{MagicWand}, a universal generation and evaluation agent that enhances prompts based on user preferences, leverages advanced generation models for high-quality content, and applies preference-aligned evaluation and refinement. In addition, we introduce \textbf{UniPreferBench}, the first large-scale benchmark with over 120K annotations for assessing user preference alignment across diverse AIGC tasks. Experiments on UniPreferBench demonstrate that MagicWand consistently generates content and evaluations that are well aligned with user preferences across a wide range of scenarios.

Figures (2)

• Figure 1: Comparison of existing AIGC methods and our MagicWand in application.
• Figure 2: Overview of our MagicWand.