by Scott Andrew
20 January 2026
In the Spring semester of 2023, I designed and implemented a Gen AI animation class that covered narrative and experimental animations. Throughout the course, students gained experience using generative AI tools as collaborators to create animations, especially during creative editing and stylization decisions.
At the beginning of the semester, I gave the students an animation project, where they chose an animation they had previously made in a prior class without Gen AI tools, such as 3D, digitally drawn, stop motion, and After Effects video composites, and then used Gen AI tools to recreate or make a new version of the project with AI tools like Deforum Stable Diffusion and Runway.ml through various processes.
What you see in the video is a side-by-side comparison of the original animations without Gen AI, and the recreated animations with Gen AI. Some of these projects went through a style transfer process that relied on their previous animation and knowledge but that resulted in the overall aesthetics of the project varying greatly. Others created much more elaborate, extended, or completely visually unique animations that centered on the existent themes in the original animations, while fully making a new work that had less reliance on style transfer techniques, including using text to video and image to video processing, and even making new reference videos that were stylized through AI.
Another goal of the class was to gauge the student’s ability to gain aesthetic and technical control over their output with AI that was at the same level of completion, skill, and control as their previous non-Gen AI projects. We also tried to gauge the students’ sense of self-efficacy as animators and generative AI users across the semester, as well as their attitudes towards ethics and IP issues with these tools.
Students used a suite of generative AI tools across all animation assignments. Student deliverables created with (treatment) and without (control) the assistance of generative AI were scored for aesthetic control, technical control, execution, resourcefulness, and fulfillment. Furthermore, changes in students’ confidence in their animation creation skills with and without the use of generative AI, and their attitudes towards AI tools were assessed at three points in the semester (January, February, March).
The results of the study suggested that at the beginning of the semester, students’ self-efficacy in creating animations without the use of AI was moderately high, whereas AI-related self-efficacy in creating animations was low. As the course progressed, AI-related self-efficacy significantly increased over the course of the semester, whereas self-efficacy in animation skills without the use of gen AI stayed the same. Concerns about AI use also changed significantly over the course of the semester. Specifically, concerns were significantly higher at the start of the course but diminished as students gained more experience using AI.
Paired-sample t-tests were conducted to test for differences in rubric scores of the AI-assisted and original animations. Rubric scores improved on each measure when AI was used; however, the difference in rubric scores between AI-assisted and original (non-AI-supported) animations was only significant for aesthetic control. Which was great to see because this was a primary hope / concern of this study. Obviously, the test group was small and the data is somewhat subjective, but I hope to use this information to craft new assignments and studies in the future that further explore these questions.