Manime: A Code-Driven Visual Teaching Method for Deep Learning Education

Abstract

This paper introduces Manime, a code-based animation teaching method for deep learning, and compares it to traditional chalk-and-board instruction using statistical analysis. While chalkboard teaching is static, Manime enables instructors to create repeatable, visually rich, and programmatically generated lessons; however, its effectiveness relative to traditional methods has not been systematically evaluated in classroom settings. We compared Manime-style instruction with chalk-and-board lectures in an introductory deep learning course and collected student feedback on comprehension, retention, engagement, and instructional preference. Using paired t-tests, one-sample t-tests, and chi-square tests on data from 60 students who experienced both formats, we found that comprehension improved by +2.77 points after the Manime animation (large effect size, d = 1.25), retention confidence was high (mean 8.28/10, very large effect size, d = 4.99), and engagement significantly favored Manime (Cramér’s V = 0.43), with students also preferring animations for difficult topics (Cramér’s V = 0.30). Students unanimously reported that visuals improved recall, and these findings align with multimedia learning theory and Dual Coding Theory, which suggest that combining visual and verbal channels enhances cognitive processing. Overall, the results indicate that Manime provides an effective and complementary teaching style to traditional chalk-and-board instruction for complex deep learning topics.