Measuring the Performance of AI-Generated Animations: Strengths, Weaknesses, and Future Directions

Generative AI has rapidly advanced in producing visually appealing animations, but its ability to capture the subtle complexities of traditional animation techniques remains insufficiently explored. This study provides a detailed evaluation of AI-generated animations through the lens of the 12 classical principles of animation, which are essential for creating lifelike and engaging motion. Utilizing expert evaluations combined with robust statistical methods, including Chi-square tests and logistic regression, we quantitatively compare AI-generated animations to traditional hand-crafted ones. Our findings reveal that while AI-generated animations perform well in static principles such as Staging, Solid Drawing, and Appeal, they consistently fail to replicate dynamic principles like Squash and Stretch, Follow-Through, and Anticipation. This analysis represents the first quantitative benchmark of AI's performance in animation, offering critical insights into its strengths and shortcomings. We conclude with specific recommendations for improving generative AI models, particularly in motion dynamics, and outline future research directions to bridge the gap between AI-driven and traditional animation techniques.