A 200-page monograph proposing a formal, falsifiable theory of cinematic greatness — and using it to analyse the arrival of generative cinema.
The thesis. A great film does not fill the viewer's time; it restructures it. We call this temporal sovereignty and derive from it a six-part criterion, stated relationally: no film is a masterpiece in itself, only for viewers in situations that permit the relation to form.
Formal core. A montage interaction kernel resolves the century-long montage debate into three coupling classes — and shows that only one, the band class descending from Griffith's continuity system, is accelerable and therefore conscriptable by an attention economy. A stimulus-inflation theorem shows that any system holding a habituating audience at constant arousal is forced onto an escalating trajectory: the virus of spectacle is a structural instability, not a moral failing. A cultivation theorem gives contemplative viewing a two-phase law with a crossover, so that the boredom of the first twenty minutes is a segment of the optimal trajectory rather than a defect. A single-carrier collapse theorem generalises density across six carriers — image, event, face, speech, sound, subject — and shows why imitation fails for a structural rather than a spiritual reason. And a single Lagrangian unifies the whole: capture and masterpiece are the same action under different weights.
Institutions. What a system can make is determined by the unit on which it levies its tax: content (the censor), the event contract (the classical studio — hence the genre bargain, which explains how the purest market system in history produced Ford, Welles, Hitchcock and Kubrick), the second (the attention market — which can fund neither), or the system as a whole (the CNC, the National Film Board). We also show how the genre bargain could be rebuilt today.
The generative turn. Documented aesthetic bias in generative models is, in our terms, an instrumentality prior. The cost of density is its inverse — and optimising models toward immediate rater approval provably narrows it. Progress, as currently defined, makes spectacle cheaper and art dearer. But the paradox is one of preference tuning, not of capability: capability is innocent, and this is a design instruction rather than a lament.
Relation to the free energy principle. Stimulus density is prediction-error rate; habituation is model learning; receptivity is precision-weighting. Read this way, stimulus inflation is an arms race against a learning model. Ontological density is defined as free energy the narrative model cannot discharge and the viewer cannot dismiss — which forces model revision, and model revision on the question why is this here is interpretation. Four frameworks with nothing in common upstream — Friston's epistemic value, Schmidhuber's compression progress, Iser's Leerstellen, and hermeneutic residue — converge on one interval.
film theory; cinematic masterpiece; temporal sovereignty; ontological density; montage; attention economy; generative AI
