Text2Movie

The film language each AI video model's docs let you write

Camera moves four tools document, in their own words

Each page takes a name people arrive with and reads that tool's own documentation for what it lets a writer ask of the camera: how a move is requested, whether shot sizes are defined, whether one generation holds several shots, and the controls around the prompt. Quoted register rows and the gathered counts sit beside the reading. As of 2026-09-25.

How one tool name becomes one page in this columnA search arrives mid-shot and names a tool. That name is read on the tool's own prompt guide, API reference or product page, which supply camera language or leave it out. The register's existing rows are quoted with their own dates, and the gathered counts are printed apart from both.A name arrivesThe tool a shot is being plannedagainstOne name, one cameraquestionRead against six fixedpointsThe tool's own docsPrompt guides, API references,product pagesA vocabulary, or silenceQuoted with the day readThe gathered materialCounts and repeated questionsTalk rather thandocumentationPrinted as what it isOne page per toolWhat each source is allowed to contribute
Fig. 1 Three sources, kept separate on every page: the tool's own words, this register's dated rows, and what the threads counted.

The searches that lead here arrive mid-shot: a path to turn into an action beat, a workflow somebody else swears by, a camera control seen in a product tour. The register answers those from maker documentation, one point at a time; the pages below answer them one tool at a time, quoting the page each line was read from and the day it was read.

Two kinds of source feed a page here and they are kept apart on it. The tool's own prompt guide, API reference or product page supplies everything about the camera, read on 2026-09-25 or on a later day for rows added since. Material gathered from one subreddit supplies the counts and the questions, kept free of usernames and of links into threads. The pipeline whose rows are quoted in each table, SceneMixer, appears by the sentence its guide published, under the same date discipline as everything else here.

Where a tool's docs say nothing, the page records the silence as silence; an unreadable reference is recorded as unreadable, which is a third state and not a judgement about the tool. Nothing on this column says how a model behaves in practice, because nothing here runs one.

1The four pages, and the tool each one reads

  • Kling's camera docs — one-take action assembled from numbered shots rather than a drawn path, read out of the multi-shot guide and the API
  • LTX Studio's camera — one name, two entries: presets picked under a prompt field, and prompt steps in the open model's guide, kept as separate rows
  • OpenArt's director — a studio page read as a studio page, with the camera vocabulary traced to the model docs underneath it
  • Runway's camera work — a workflow read after its named control was retired: prompt ingredients, an app for keyframes, and the threads' pricing questions

Each page below prints the day its sources were read, and the register's rows keep their own older dates where a reading was taken earlier. How a doc is read, and what counts as covered, is set out on the method page; the register the quoted rows come from is at the models reel.