Pre-processing as Poetics.
Modelling Film Editing as Space Puzzle
An experiment about representations of film editing, and their relation to theories of time.
Film editing creates meaning
MoNetz
Montage Networks (MoNetz) is a conceptual representation of editing relations and patterns between shots in a film — loosely inspired by the Tonnetz in music theory — a way to model and explore editing space computationally.
Editing as a puzzle board
The analogy comes from spatial state puzzles like Klotski, where the space of possible board configurations can be drawn as a graph — and the puzzle's difficulty read off the graph's shape. We represent the transitions of shot scale in popular films as positions in a board.
- State — each shot occupies a cell on a 5×5 grid defined by (previous shot scale, current shot scale): ECU, CU, MS, FS, LS.
- Transition — a cut to the next shot changes the spatial configuration to a new state.
- Local structure — within a scene, the film is likely to stay in a neighbourhood of the board (e.g. alternating MS↔CU↔MS... across a conversation).
- Global structure — scene changes are more likely to jump to distant cells.
The film's edit sequence traces a path through this 25-cell state space, similar to how a puzzle solver traces a path through configuration space.
Three case studies
Three films, three editing economies — a classical Hollywood thriller, a contemporary Hollywood blockbuster, and Soviet-era long-take art cinema. The figures that follow trace each one's edit sequence through the same 5×5 state grid.
Vertigo (1958) ↗
Alfred Hitchcock's psychological thriller of obsession and vertiginous descent, built on tightly controlled point-of-view editing.
Editing style — economical, continuity-driven découpage: shot/reverse-shot conversations alternate closely between CU and MS, punctuated by wider establishing shots at scene changes.
Inception (2010) ↗
Christopher Nolan's heist film nested across dream layers, cross-cutting between simultaneous timelines and levels of reality.
Editing style — high shot count and rapid alternation across scale registers within a scene, plus frequent parallel-action cross-cutting between layers: a denser, more agitated path across the grid.
Stalker (1979) ↗
Andrei Tarkovsky's contemplative journey through the Zone, one of Soviet cinema's defining works of slow, durational filmmaking.
Editing style — sparse cuts, long-held wide and medium shots, minimal shot/reverse-shot alternation: a sparse, wide-ranging path with long dwell times per state.
Visualisation 1
Edit paths on the state grid
Tracing a film's actual edit sequence as a path on the grid, coloured by time (early → late), reveals whether the film lingers in a local neighbourhood or jumps restlessly across scale registers.
Side by side, the two paths already hint at different editing economies — one denser and more clustered, the other sparser and more wide-ranging.
Visualisation 2
Temporal neighbourhood map
For each film, we measure the Manhattan distance between consecutive states on the grid, plotted over the edit sequence. Low values mean the film is staying in a local neighbourhood (within a scene); spikes mark scene or sequence changes. This is designed to reveal the rhythm of local versus global structure.
Visualisation 3 — Interactive
Watching the montage network build
Each animation plays a film's edit sequence shot by shot: a node grows (or appears) each time the film revisits a bigram state, an edge appears for each state-to-state transition, and the current shot's thumbnail is shown alongside its scale label. Node size is weighted by cumulative screen time, not just visit count — so scenes the film lingers in visibly dominate the board.
Press ▶ Play, or drag the slider to step through transitions manually. Hover any node for details.
Theories of time
The representations shown above are underpinned by a conceptualisation of time: what gets counted is an expression of a temporal form before it becomes formalised. What shape should we give time?
Time as a river
Three dimensions + change
Time as a landscape
Four dimensions (no change)
Ongoing and connected research
MoNetz and its various experiments are part of the ongoing Sculpting Time with Computers proof of concept, developed in collaboration with King's Digital Lab.
A companion tool for shot-level visual analysis, developed with King's Digital Lab.
A related King's project this work draws on and feeds back into.
Get in touch
