What the prompt cannot teach: textual mediation and the limits of generative video in film education

This article draws on two pedagogical interventions in which twenty-seven students at a small public university used generative video AI to produce short films. This observational study examines how students engaged with these tools and what pedagogical possibilities and constraints they encountered. Some constraints are conjunctural; others, I argue, are structural: they share a single root, the textual mediation through which these models operate, at training and generation alike. That mediation renders poorly all that is non-verbal in filmmaking – tacit knowledge, embodied experience – the very things students should begin to acquire at university. Students are thus left with no way to use their instinct or prior experience, and no choice but to validate whatever the model returns, revealing what I call a ‘threshold of acceptability': a validation by frustration or resignation that marks a pedagogical dead end. The first intervention exposes these limits; the second, confining AI to a post-shoot video-to-video transformation, points to uses where practical learning is preserved. I conclude that the deeper risk is that filmmaking education may undergo a reverse adaptation to what AI tools allow – which makes critical media education more necessary than ever, though it cannot replace the lost practical education.