Docs / 04

Directors and films: casting by dramatic problem

How Cinematic UI casts a director by the product's dramatic problem, extracts testable mechanisms, and researches any film with sourced, tagged claims.

A director is a set of habits about where the camera stands, how long a shot lasts, what the light hides and how a cut lands. Cinematic UI hires those habits for a product. The casting book in references/directors.md lists about forty directors as mechanisms, each with the obvious pick to avoid and the lazy version of using them. It is a starting shelf, not a limit.

Casting by dramatic problem

The first question is not “what genre is this brand?” but “what must the audience feel that they do not feel on arrival?” Trust, awe, desire, calm, urgency, delight, belonging, clarity. The skill casts for that, not for the industry.

The method, in short:

1. Name the dramatic problem the product has.
2. Write down the three obvious picks and set them aside. For tech these are Villeneuve, Nolan, Kubrick. For lifestyle, Wes Anderson and Sofia Coppola. For night and fashion, Wong Kar-wai and Refn. Every model reaches for them, so a site cast that way looks like every other site.
3. Cast against type when it serves, as long as the treatment makes the product clearer and not stranger.
4. One director, one film. The film narrows the director to a period, palette and rhythm. You may borrow one department from someone else and say so, but a blend of three directors is a mood board.
5. Extract five to seven mechanisms, covering at least camera, light and colour, editing, and one of staging, type or sound.
6. Hold a casting session: three candidates, one kept, reasons recorded, and the filmography checked so the next production differs from the last three.

If you name a director or film, yours wins. The agent’s job is then the same: mechanisms, not costume.

The mechanisms test

Every mechanism is one concrete, observable decision plus its translation into this site. Two questions decide whether it counts:

• Could a developer build it from this sentence?
• Could a viewer point at it on screen?

The treatments of the demo productions show what this looks like. For the AI operations platform Lockwork, the cast film is Robert Wise’s The Andromeda Strain (1971). One mechanism is “the descent through levels, a door at each”, which becomes a lift: four floor buttons, and each press closes two door halves over the stage and opens on the next level, with the ground stepping one tone darker. Another is “the one man with the key”, which becomes a gate where every irreversible step waits for a named person. Neither uses a prop or a quote from the film.

A final check on the winner: could a competitor use this direction unchanged? Lockwork’s four levels, its key-plate buttons and its diff-as-type-switch come from its own permission model, so a generic agent site could not borrow them.

Any film can be cast

The skill has no film library. When you name a film, it researches it. When you describe a look instead, for example “90s Hong Kong late night, lonely, humid”, it searches for films that match the picture, not only the theme, and offers candidates. Either way it follows references/film-research.md:

• Research runs in capped rounds: credits, light and grade, art direction, poster and titles, plus searches in the film’s own language.
• Every line is tagged for how it is known.

• Reference stills, set details and the poster are looked at before anything is built, but never shipped. They are reference only.
• With no web access, the skill falls back honestly and marks what it could not source.

The fourteen looks in the grade book and the worked examples in the art-direction book are examples of this process, not a menu. A lazy cut the skill refuses is “a film remembered instead of researched”: credits, palettes and set details written from memory and presented as fact.