Interactive filmmaking: a director’s FMV authoring workflow

The first decision is not the plot and not the cast — it is the structure: how the branches fork and rejoin. The nodal scheme (fork → shot branches → rejoin into the shared flow) is the cheapest and the oldest — Kinoautomat ran on it. Parallel, labyrinth and environment-based structures cost differently, and the difference multiplies with every scene. I choose the structure before the script and budget the branches from it: the branching-structures guide is the frame I start from.

A script for interactive filmmaking is not “a film with forks” — it is a graph. I number the nodes (the scene where a choice happens → the consequence scenes) and for each node write down three things: what the viewer sees before the choice, which options exist, and where each option leads. The money-saving rule: every branch must rejoin somewhere. A tree without rejoin points multiplies the shoot; a graph with hubs does not. The format can be a spreadsheet or Twine — but the node numbering has to survive onto the set: it becomes the crew’s language.

Continuity

Continuity here is not “one scene, one day” — it is agreement between branches the viewer may see in any combination. Costume, hair, props and light level are locked for the whole block of branches, not for a scene. The cheapest insurance is shooting all branches of one node as one block while the set stands.

Idle loops

While the choice timer runs or the branch loads, something must be happening on screen: short loops — a glance, a breath, a hand on the door handle — that cycle without a visible seam. They cost minutes on set and save the rhythm in the build. Without them an interactive film “clicks” at every choice.

Coverage for transitions

Every branch-to-branch transition needs an edit bridge: a reaction shot, a cutaway, a detail. I shoot cutaways at every node even when the script says they will not be needed — travelling back for a missing cutaway costs more than one extra setup.

The edit has two layers: inside a scene — ordinary cinematic grammar; between scenes — node editing, where the exit of one branch must match the entrance of any next one. The working method: cut the “spine” (the most likely path) first, then verify every fork at its seams. Sound bridges — ambience that continues across the seam, a musical pedal — make transitions seamless cheaper than any video trick.

Authoring is when the graph becomes playable. Three routes I know, fastest to most flexible.

Stornaway

A browser tool without code: clips become “islands”, paths are drawn with the mouse, and web is the standard export. The case I read is Paul Raschid on Stornaway: scripts as playable gamebooks, the story map on set, post-production updating a living graph. For a web film without a programmer, it is the shortest path.

Unity VideoPlayer

The engine buys everything a video switcher cannot do: hotspots on stills, QTE with timing curves, items, state across scenes, export to PC and consoles. The technical base is VideoPlayer: video as a scene component with frame control and events. This is the FMV-game route, with the engineering bill that comes with it.

FMV Builder

My own tool: a studio for interactive films and visual novels — a branching editor, FMV mechanics (QTE, hold, explore, items, dialogue trees), a timeline, and export to web or an offline build. Disclosure: it is my project, so my words about it are not an independent review. The comparison with alternatives, conflict stated, is in the tools guide; the development status is on the FMV Builder page.

Testing a branching film is not “playing the game”: it is walking every critical path of the graph and checking the seams. My minimum: a path matrix (which choice combinations must be walked), a run of every ending, and a separate look at the idle loops — they break first. Release is a web build or a store page; sitemap and metadata for the web version are covered in the FMV-01 devlog.

The six steps above are how I build FMV-01: a cyberpunk-noir prototype with video cutscenes, stills with hotspots, and a hire-or-walk fork. The structure is nodal; the script lives as a graph; the shoot runs in branch blocks with idle loops; the authoring is my own FMV Builder. What already works and what is in development is honestly stated on the project page and in the devlog. What it costs and how long it takes: there are no public numbers yet — the devlog documents the pipeline, not the budget; I will publish figures with the devlog rather than invent them here.