Video-based learning is becoming a popular platform where learning can happen in various life-related domains. However, learners may lack prior knowledge needed to select appropriate videos and may struggle to identify and link key parts in those videos, leading to frustration and feeling overwhelmed. To address this challenge, we present a novel framework for generating Video Narratives for Learning (VIN-L). A video narrative is a combination of video segments which have been automatically characterised and linked by computer algorithms. VIN-L generates video narratives by using a domain taxonomy to link characterised segments from different videos following Ausubel’s Subsumption theory for meaningful learning. Four video narrative types have been generated implementing subsumption links: Derivative, Super Ordinate, Correlative, and Combinational. VIN-L has been applied on a corpus of 60 videos related to health quality of life needs of people living with chronic respiratory illnesses. The resulting video narratives have been evaluated with domain experts and with Amazon Mechanical Turk workers, assessing quality, perceived usefulness, and the potential of using the video narratives to support learning. Domain experts’ results show that the generated video narratives are of good quality and have the potential to support learning. The Amazon Mechanical Turk study has shown that the generated video narratives have potential to be useful for learning, helping with identifying and linking key domain terms.