localgpt.md previs
Drop a screenplay — a PDF, a Final Draft file, Fountain, or script pages in Markdown — and get a previs board: a set, a stand-in per speaking character and a first pass of coverage — a master and a single on everyone — with real lenses and measured camera positions. It happens in this tab. No install, no account, no upload: the script never leaves your browser.
Drop a
.pdf, .fdx, .fountain or .md file hereor click to choose one · the script's PDF, a Final Draft file as saved (
.fdx), Fountain from Highland, Beat, Slugline or any export — or Markdown pages: ## INT. KITCHEN - DAY is a scene, **MAYA:** opens a lineNo script handy? Try the sample as a , as , or
Loading the draft…
What you get
- Per scene, a
.worldpackage: the set (a room forINT., a location under a sun forEXT.), the cast as stand-ins, and the coverage asext-cinematographycameras — a 24 mm master cropped to 2.39 and a 50 mm over on each speaker, aimed and focused at the actor by entity, so a revision that moves the actor keeps the shot. - A shot list (
shotlist.csv): scene, shot, size, lens, sensor, aspect, camera height, distance to subject, in and out on the clock. - A printable board (
board.html): one card per shot with its frame rendered by the Open World Format's reference viewer — the same file this page shows below. - The draft is deterministic and model-free: the same pages give the same board, byte for byte,
here and in the
localgpt previscommand line. Revising shots by asking — "make 1B a 35", "stay on her", a treatment as a branch — is the desktop app's half: unzip the packages and open one scene live,localgpt-gen --live scene-1.world, to look through the shots and put an agent on the folder (the walk).
PDFs
- A screenplay page is a fixed layout, and that is how a PDF is read: the text layer's lines by where
they sit — action and scene headings at the left margin, dialogue an inch in, cues further still,
transitions at the right — with the columns learned from the file, since every app sets them a few
points apart. Scene numbers in the margins stay with their headings; page numbers,
(CONTINUED)and(MORE)go. A Fountain or plain-text file printed as it is, all at one margin, is read as Fountain. - The page's lines are the script's lines, so the page-time estimate reads a PDF the way a page is read. A scanned script has no text to read and is refused as one — export the text, or OCR it first. Dual dialogue set side by side comes out as one cue. Up to 300 pages.
Final Draft files
- A
.fdxis read as Final Draft saved it: scene headings with their scene numbers, characters, parentheticals, dialogue (dual dialogue too), transitions, action. Acts become sections, script notes become notes, the cast list is dropped. The title page's first line is the title; a "written by" line and the one after it are the credit and the author. - Final Draft wraps action by its page layout, so the draft re-wraps it at a screenplay page's 61-character action column before timing it — the same page-time estimate a hard-wrapped Fountain file gets. Styling, revision marks and page breaks don't reach the board. The four samples above are the same screenplay four ways, and they draft to the same board, byte for byte.
Script pages in Markdown
- A heading at any depth that reads as a scene heading —
## INT. KITCHEN - DAY,### EXT. CLIFF PATH - DUSK— is that scene. Any other heading is an act or sequence. - A bold name with a colon opens dialogue:
**MAYA:** You're late.A leading*(beat)*is the parenthetical. A plainMAYA: You're late.works when the name is uppercase. - Front matter (
title:,author:,credit:) is the title page; failing that, the first#heading before any scene is the title. An HTML comment is a note. - Everything else is action, as written — and Fountain inside a
.mdfile passes through untouched: an uppercase cue over its lines, a bareINT.line,> CUT TO:.