BRENNEN

Industries / Media, arts and recreation / Media and entertainment / Film production

Preparation

Preparation, or pre-production, is the work between a green-lit script and the first day of shooting: breaking the script into lists, scheduling and budgeting the shoot, casting, finding and clearing locations, drawing or animating shots in advance, designing sets and costumes, and planning for safety and intimate scenes.

7

Milestones achieved, from Early 1930s to 2019.

3

To watch for: announced, not yet confirmed outside.

3

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. Early 1930s

    Stories pinned to a wall

    At the Walt Disney Studio, story artists began pinning sketches in order on 4-by-8-foot boards, so a film could be planned from pictures instead of read from a script. Disney's archive says the method spread across the film industry, "especially for filmmakers planning commercials and action sequences of live-action films," and in 2019 the director Jon Favreau said The Lion King started "with storyboards."

    d23.comd23.comdga.org

    Credit

    Webb Smith, a Disney story man from 1931, according to Disney's archive; the credit is disputed (see Checks)

  2. 1971

    Casting calls get a clearinghouse

    Gary Marsh started Breakdown Services, which writes breakdowns, short descriptions of each role in a script, and sends them to talent agents and managers, who put actors forward to casting directors. In 2006 Backstage called breakdowns "the major medium" through which casting directors, agents, managers and performers find one another, and by 2011 the company reported affiliates in the UK, Australia and New Zealand.

    breakdownservices.combackstage.comprnewswire.com

    Credit

    Gary Marsh, Breakdown Services

  3. 1976

    War footage as a guide

    Before the space battles of Star Wars were filmed, George Lucas cut World War II air-combat footage into a sequence and gave it to the effects crew at Industrial Light & Magic (ILM) as a guide; Lucasfilm calls it "a sort of proto-animatic," an animatic being a rough moving version of a storyboard. While the battle was edited in fall 1976, Marcia Lucas cut in dogfight footage wherever effects shots were not yet ready. The practice grew into previsualization, or previs: planning shots as rough animation before they are filmed.

    lucasfilm.comexhibits.library.illinois.edu

    Credit

    George Lucas; Marcia Lucas; ILM

  4. 1985

    The strip board goes digital

    Screenplay Systems, the Burbank company of Stephen Greenfield and Chris Huntley, released Movie Magic Budgeting in 1985 and Movie Magic Scheduling/Breakdown in 1987. The scheduling program put the stripboard into software. A stripboard is a board of colored strips, one per scene, shuffled into the order the scenes will be shot, and the software let the schedule flow into the budget. A London film school's course page still calls the pair "the recognised industry standard."

    screenplay.comhumanitasprize.orglondonfilmacademy.com

    Credit

    Stephen Greenfield and Chris Huntley, Screenplay Systems

  5. 2014

    A refused location, used anyway

    On 20 February 2014 a freight train hit the crew of Midnight Rider on a railroad trestle over Georgia's Altamaha River, killing the camera assistant Sarah Jones, 27, and injuring eight other workers, by OSHA's count. OSHA, the US workplace safety regulator, put the failure in planning: the company had no safety plan and no permission from the rail owner to use the tracks, and the railroad, CSX, said it had twice refused. The director Randall Miller pleaded guilty to involuntary manslaughter and criminal trespass in March 2015 and was sentenced to two years in county jail and eight years' probation.

    LessonPermission to use a location is part of the safety plan, and when the owner refuses, the scene has to be replanned.

    ehstoday.comweb.archive.orgcbsnews.com

    Credit

    OSHA; Georgia prosecutors

  6. 2018

    Intimate scenes planned in advance

    In October 2018 HBO began requiring an intimacy coordinator, who plans and choreographs sex scenes and checks each actor's consent, on every show with a sex scene; the rule followed the actor Emily Meade's request for one on The Deuce. In January 2020 the actors' union SAG-AFTRA published standards saying the coordinator meets the producer or writer and the director before filming to settle the degree of nudity and the details of simulated sex. The union did not make coordinators mandatory.

    avclub.compastemagazine.comkqed.org

    Credit

    HBO; Emily Meade; SAG-AFTRA

  7. 2019

    Scouting sets in headsets

    For The Lion King, Jon Favreau's team built every set first in a game engine, software made to draw 3D worlds in real time, and then scouted the sets together in virtual reality, about six people in headsets including the cinematographer Caleb Deschanel and the production designer James Chinlund, before filming with a full crew on an empty stage.

    dga.org

    Credit

    Jon Favreau and his team, by Favreau's account in the Directors Guild's magazine

Watch for

  1. Breakdowns in under an hour

    Dramatify, a software company in Stockholm, says its AI Script Breakdown finds the cast, props, wardrobe, vehicles, animals, stunts, locations and effects in a screenplay and produces a "production-ready breakdown" in under an hour, a job that "can take days"; its chief executive says it cuts "80 per cent of the grunt work."

    Achieved whenSomeone outside Dramatify runs it on produced screenplays and publishes how many elements it found and missed against an assistant director's breakdown, or a production publishes the hours it measured as saved.

    broadcastnow.co.uk

    Claimed by

    Dramatify (chief executive Annika Lidne), June 2025

  2. Generative AI in Netflix previs

    Netflix says it used generative AI on roughly 300 titles in 2026 by July, mostly in post-production, and that its use runs "from concept development and previsualization through delivery."

    Achieved whenNamed titles are shown to have used generative AI for previs, confirmed by people who worked on them or by an outside report that says what it replaced.

    tvbeurope.comthe-decoder.com

    Claimed by

    Netflix, in its July 2026 earnings report and on the call with co-CEO Ted Sarandos

  3. World models for prototyping scenes

    Google DeepMind released Genie 4, a world model that draws a scene a person can move through from a typed description (see the Note "A world that follows"), as a research preview for trusted testers. Testers point to prototyping films, a DeepMind engineer said its real-time response matters most "in the storyboarding phase," and Google says a scene stays consistent for up to two minutes.

    Achieved whenA production outside Google shows scouting or storyboard material made with Genie that it used to plan a real shoot, confirmed by its crew or the trade press.

    techcrunch.com

    Claimed by

    Google DeepMind, October 2026

Open

  1. 01

    Score AI breakdowns against ADs

    A public benchmark. Take 20 screenplays of produced films or shorts whose rights holders allow it. Two working assistant directors (ADs) each break every script down on their own into the standard lists: cast, extras, props, wardrobe, vehicles, animals, stunts, effects and locations. Run the same scripts through AI tools and score each tool's lists against the ADs', category by category, with the agreement between the two ADs as the ceiling.

    The brief
    Credit · unclaimed
  2. 02

    AI schedules against real ones

    Collect finished shooting schedules from 10 completed student, independent or short-film productions, with permission and with names replaced by cast numbers. Give each tool the same breakdown and the limits the AD faced, such as actor availability, location dates and day or night, and compare each tool's schedule with the one the production actually shot.

    The brief
    Credit · unclaimed
  3. 03

    Map film permit rules worldwide

    An open, machine-readable table of filming permit rules for 30 major filming cities on at least five continents: permit fee, advance notice, insurance minimum, notice to residents, police or fire requirements, drone rules, permitted hours, office contact, source link and date checked. Each film office is then asked to confirm or correct its row.

    The brief
    Credit · unclaimed

Open 01 · the brief

Score AI breakdowns against ADs

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public benchmark. Take 20 screenplays of produced films or shorts whose rights holders allow it. Two working assistant directors (ADs) each break every script down on their own into the standard lists: cast, extras, props, wardrobe, vehicles, animals, stunts, effects and locations. Run the same scripts through AI tools and score each tool's lists against the ADs', category by category, with the agreement between the two ADs as the ceiling.

Why it is still open

The public evidence is one-off. A 2023 hands-on review of Filmustage on one feature script said it "generally didn't miss any elements" but gave no tally; a September 2026 research paper, PACE, measured breakdown coverage on one screenplay against a reference built by a model. This research found no reference set made by ADs.

noamkroll.comarxiv.org

Built with what exists

Filmustage and Dramatify, two commercial AI breakdown tools; general models such as Claude Fable 5.2 (see the Note "Three working devices") and Meta's Muse Ember for extraction (see the Note "The jobs agents repeat"); Gemini Decision 1 to judge whether a tool's item matches an AD's item (see the Note "A thousand questions, one pass"), with a person spot-checking its judgments.

  1. First step

    Write the category rules and the scoring sheet, have one AD break down one short screenplay, and run it through two tools.

  2. Done looks like

    20 scripts, two AD breakdowns each, at least four tools, and a published table per category of precision (the share of a tool's items the ADs also listed) and recall (the share of the ADs' items the tool found), beside the AD-to-AD agreement.

Open 02 · the brief

AI schedules against real ones

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

Collect finished shooting schedules from 10 completed student, independent or short-film productions, with permission and with names replaced by cast numbers. Give each tool the same breakdown and the limits the AD faced, such as actor availability, location dates and day or night, and compare each tool's schedule with the one the production actually shot.

Why it is still open

Research on film scheduling going back at least to 1993 tested its methods on "randomly generated problem instances," not real schedules. Filmustage now sells an AI agent that "optimizes shooting days," and its March 2026 article describes the agent through the company's own product materials and webinars, not a measured test. This research found no published comparison with real productions' schedules.

link.springer.comfilmustage.comfilmustage.com

Built with what exists

Google OR-Tools, open-source optimization software, as a baseline without AI; Filmustage's AI production agent; a general model such as Claude Fable 5.2 (see the Note "Three working devices"); a spreadsheet for scoring.

developers.google.com

  1. First step

    Ask three film schools or independent producers for one finished schedule each, and build the scoring sheet on the first one.

  2. Done looks like

    For each of the 10 productions, a published table comparing the AD's schedule and each tool's on shoot days, company moves (moving the whole unit to a new location), cast hold days (days an actor is paid but not filming) and broken limits, plus a blind ranking of the schedules by three ADs.

Open 03 · the brief

Map film permit rules worldwide

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open, machine-readable table of filming permit rules for 30 major filming cities on at least five continents: permit fee, advance notice, insurance minimum, notice to residents, police or fire requirements, drone rules, permitted hours, office contact, source link and date checked. Each film office is then asked to confirm or correct its row.

Why it is still open

The best regional compendium found, the California Film Commission's 30-Mile Studio Zone Chart, is a 52-page PDF whose August 2026 version is marked "update in progess." A 2026 Giggster report compares one standard permit fee across 40 US markets, not the rules. This research found no worldwide set that software can read and that the offices have checked.

film.ca.govgiggster.com

Built with what exists

Browser Use 1 to visit film office pages (see the Note "Deleting 80% to start over"); Muse Ember to pull fields out of PDFs and long pages (see the Note "The jobs agents repeat"); Claude to check rows for contradictions (see the Note "Seven more flagged"); a CSV file in a public repository.

  1. First step

    Pick the 30 cities and fill in three by hand to settle the rules for each field.

  2. Done looks like

    30 cities with 10 fields each, every cell linked to an official page; written replies from at least 15 offices; and a published count of the cells the offices corrected, which serves as the agent's error rate.