research · 2025
Generative Lecture: Making Lecture Videos Interactive with LLMs and AI Clone Instructors
arXiv
@misc{jo2025generativelecturemakinglecture,
title = {Generative Lecture: Making Lecture Videos Interactive with LLMs and AI Clone Instructors},
author = {Hye-Young Jo and Ada Yi Zhao and Xiaoan Liu and Ryo Suzuki},
year = {2025},
eprint = {2512.21796},
archivePrefix = {arXiv},
primaryClass = {cs.HC},
url = {https://arxiv.org/abs/2512.21796}
}Generative Lecture embeds AI clone instructors into existing lecture videos using LLMs and avatar/voice synthesis, adding on-demand clarification, interactive examples, adaptive quizzes, and personalized explanations. A study (N=12) showed it supports personalized learning.
Problem: lecture videos are the same for everyone
Lecture videos on platforms like Coursera and Khan Academy are static: once published, every viewer watches exactly the same content, regardless of their needs or context. The experience is passive and one-directional: unlike an in-person lecture, you can’t ask the video a question.
Solution: an AI clone instructor that answers inside the video
Generative Lecture turns an existing lecture video into an interactive, personalized experience. An AI clone of the instructor (built with HeyGen, ElevenLabs, and GPT-5) is embedded into the video, and the system generates new material on demand in response to a student’s question, inserting it into the slides so the instructor appears to answer directly.
From a design elicitation study (N=8) we identified four goals, which guided eight features: on-demand clarification, enhanced visuals, interactive examples, personalized explanations, adaptive quizzes, study summaries, automatic highlights, and adaptive breaks.
Evaluation
A user study (N=12) and expert feedback (N=5) suggested the system supports effective two-way communication and personalized learning.