Marquee Studio

The principle

Marquee's AI may notice, structure, and surround the writer's words — it may not compose, replace, or paraphrase them.

A binding commitment, published here so it can be held to.

The promise

Marquee will not write your screenplay. Every word of every action line, every line of dialogue, every parenthetical on every page is yours — typed by you, owned by you, WGA-clean.

Marquee notices things: that you've used V.O. for JUNE eleven times and O.S. once, that your inciting incident lands on page 14, that DETECTIVE NOLAN and DAD are probably the same character.

Marquee fills in the structure around your words: scene headings from locations you've already used, character cues that already exist in your script, a title page from your metadata.

Marquee never finishes your sentence, tightens your dialogue, fixes a typo in your action, or drafts the next beat. There are tools that do those things. This isn't one of them.

The four levels

Four kinds of AI participation. We do three of them.

  1. L1

    Marquee does this

    Structural notice

    Margin notes, Bible entries, scores. Output that lives beside the script, never inside it.

    Always in.

  2. L2

    Marquee does this

    Form completion

    Deterministic ALL-CAPS tokens — sluglines, character cues, transitions, V.O./O.S./CONT'D. Pulled from what you already wrote.

    In, retrieval only. Never LLM-generated.

  3. L3

    Marquee does this

    Content observation

    Craft notes about what is already on the page — drift, consistency, beat-structure gaps, coverage. Always in the margin, always with a provenance chip.

    In, in margin only.

  4. L4

    Marquee does not

    Prose generation

    Drafting, expanding, rewriting, tightening, paraphrasing action, dialogue, parentheticals.

    Always out.

The twelve cases

What Marquee will and won't do, line by line.

Twelve specific features, decided. Decided means they don't get relitigated when a product manager has a clever idea on a Tuesday.

01

Autocomplete a slugline you've already used (INT. KITCHEN — DAY)

L2 — retrieval from your own headings.

IN
02

Autocomplete a character name above a dialogue block

L2 — only names that already exist in the script.

IN
03

Fix a typo in an action line

OS spellcheck is fine. Marquee does not touch action.

OUT
04

Fix a typo in dialogue

Voice is the asset. We leave it alone.

OUT
05

Tighter rewrite of an action line

A margin note is welcome. A rewrite never is.

OUT
06

Tighter rewrite of dialogue

Same rule. The line stays yours.

OUT
07

Generate a logline

Metadata, not script body. Writer edits required before export.

IN
08

Generate a synopsis

With friction — required edit step. Variety found AI synopses read like 11th-grade essays.

IN
09

Generate a title page

Pure metadata composition.

IN
10

Beat-structure observation

Bible or margin only. Page numbers and structural diagnostics, not prose.

IN
11

Suggest the next plot beat

In as observation ("38 pages between inciting and midpoint"). Out if it names the beat ("try: JUNE discovers the letter").

GRAY
12

Auto-format a Fountain paste

Structural parse. No new characters, no new lines.

IN

The gray-area rule

How we decide what we haven't decided yet.

New ideas land all the time. When one does, we walk it through these four questions, in order, and stop at the first one that resolves it.

  1. 1

    Does the output land inside a slugline-bounded scene? If yes, it must be L2 or it is out.

  2. 2

    Would the writer's own words be replaced, paraphrased, or composed with? If yes, it is out.

  3. 3

    Is the surface the margin, the Bible, the score, or some other separate metadata? If yes, L1 or L3 is in — with a provenance chip and a writer-must-accept step.

  4. 4

    Could a Warner Bros story analyst do this exact thing? If yes, it is in.

The writer's words are sacred; the surface around them is ours.