Interactive Cinema: Theory, Practice, and AI Methodologies

The course consists of four content blocks (eight lectures) and practical studios; the table below shows the structure by blocks rather than weeks — the calendar depends on the semester schedule. Every block pairs theory with one practical assignment, and the studios lead from deconstructing an existing work to a production package of your own.

By the end of the course each student has three artefacts: a deconstruction of a chosen interactive work (a written breakdown of its choice structure), a working prototype of branching logic, and a completed production package with a pitch. Together they form a portfolio one can show a producer or a department — and it is assessed as a system of decisions, not a set of generated images.

Generative AI has two working roles in the course. The first is research: AI-assisted study of the genre’s history, where reconstructing lost works from textual sources teaches you to check every model claim against a primary source. The second is production: concept art and visual materials for students’ own projects, with mandatory disclosure of AI involvement.

The narrative logic itself is made not by models but by the author: branching scenarios are mapped in Twine and Arcweave, and timed-choice interfaces are prototyped in React. AI helps research and illustrate in this chain, but it does not make dramaturgical decisions — that is the course’s core rule, carried through every block.

The course’s practical case is Vasya the Reformer, Dovzhenko’s lost 1926 debut, which I reconstruct from surviving textual sources: the scripts are broken into a forty-scene storyboard, characters are fixed in character plates, and frames are generated scene by scene — each linked to its place in the text. For students it is a live example of “reconstruction from text”: the hierarchy of trust in sources, the boundary between interpretation and invention, honest labelling of AI frames. The methodology and the frame gallery are on the project page.

The final assignment is a production package: a branching scenario, a logic prototype, concept art, montage, and a pitch defence. What is assessed is the decision, not the pixels: how justified the branching structure is, whether choices carry consequences, whether the style holds together — regardless of whether the material was generated or hand-made. Every use of generative AI in the work must be disclosed.

The course’s ethical rules run through every assignment. First: AI involvement is always labelled — in concept art, reconstruction frames, and texts. Second: archival materials are a primary source, not raw material for free fantasy; reconstruction builds on surviving scripts and filmographic cards. Third: an AI frame of a lost film is an interpretation, not a “restoration of the footage” — and the wording matters; the Vasya case is where we practise it.

The disclosure format is simple: a caption under the material — “generated from the description of scene N”, “an AI concept based on source X”. This is the same practice I use in my own open projects, and students carry it into their own public work: a label does not devalue the work, it makes it verifiable.

Ukraine has commercial programmes on AI in film: Filmschool’s “AI in Cinema” course, Robot Dreams’ “AI for Video Production”, and Jemads’ materials on cinematic AI. They focus mostly on production tools. The university module differs in three ways: it carries ECTS credits inside an audiovisual-art programme; it teaches the theory of interactive cinema — history, typology, an analysis model — rather than tool skills alone; and it runs on an open methodology with a real research case that can be verified on this site.

The instructor is Sasha Poberailo, a filmmaker and researcher of interactive cinema, lecturer at the Kharkiv State Academy of Culture (more about the author).