The Role of AI in Reshaping Our Understanding of Narrative Performance

Text-to-video generative artificial intelligence models can now produce compelling visual results of characters interacting with their environments across a broad range of artistic/filmic styles. However, these portrayals of characters should not be confused with narrative performances: the semantic models used to produce such videos lack the ability to semantically understand their own results. This presentation explores why meta-understanding is a crucial component of narrative performance for both philosophical and practical considerations. In particular, the approaches used by current video generation models (VGMs) do not provide artists with the means to direct the resulting imagery without fully re-rendering a shot, which violates the understanding that restored behaviour is a necessary element of performance. This presentation demonstrates how specific types of outputs from VGMs will be essential for serious adoption by filmmakers and animators, and how this relates to our understanding of what constitutes narrative performance.