In this article
AI is useful in pre-production for clerical first passes: tagging a script breakdown, summarising scenes, drafting documents and sketching previs frames. It is not good at judgment, at knowing your locations and people, or at being accountable. Use it to propose, have a named person check and decide, and keep it away from the creative calls that make the film yours, including the final cut.
Key takeaways
- AI is strongest on repetitive, checkable work: tagging, summarising, formatting, first drafts.
- It is weakest where a decision needs context, judgment or accountability.
- Every AI suggestion needs a named person who confirms it before it reaches a department.
- Treat unreleased scripts and actors' likenesses as confidential; know where your data goes and get consent.
- AI proposes, people decide. It should never make the creative decisions, and never touch the final cut.
Most of the noise about AI in film is about generating images and video. In pre-production the more useful questions are smaller and more practical: which hours of clerical work can a tool take off a first AD or a producer, how do you check what it produced, and which decisions should never be handed to it at all.
What is AI actually good at in pre-production?
The common thread is work that is repetitive, has a clear right answer most of the time, and is easy for a person to check.
- First-pass script breakdowns. Reading every scene and suggesting cast, props, wardrobe, vehicles and effects is exactly the kind of tedious, systematic task where a tool saves real time. A person still reviews every tag.
- Summaries. One-line scene synopses for the stripboard, character summaries for casting, or a quick digest of a long notes thread.
- Formatting and conversion. Turning a messy draft into standard screenplay format, or a list of requirements into a tidy department sheet.
- First drafts of documents. A crew brief, a location request letter, a deal memo outline. Useful as a starting point, never as the final text.
- Previs and storyboard frames. Rough frames help a director communicate a shot to the DP and the designer before anyone has drawn a line. They are a sketch, not a design.
- Translation. A first pass at translating a call sheet or a schedule note for a multilingual crew, checked by someone fluent.
What is AI not good at?
- Judgment. Whether a scene needs two days or one, whether a performance is working, whether to cut a subplot to save a location: these depend on taste, experience and the people involved.
- Knowing your world. A tool does not know that the location floods at high tide, that an actor is unavailable on Fridays, or that two HODs do not work well together. You do.
- Safety. Stunts, water, height, fire, weapons, animals and children need qualified people making decisions on the day. A suggestion from a model is not a safety plan.
- Being right without checking. Language models can produce confident, plausible and wrong answers, including details that do not exist in the script. Assume errors and look for them.
- Accountability. When something goes wrong, someone has to own the decision. A tool cannot.
AI can make the first draft faster. It cannot decide what the film is.
How do you keep people in charge of AI suggestions?
Write the rules down before the tools are in daily use, so nobody has to improvise them under deadline pressure.
- Proposals, not changes. AI output should arrive as a suggestion a person accepts or rejects, never as a silent edit to the breakdown, schedule or budget.
- A named reviewer for every output. The first AD confirms breakdown tags; the producer confirms anything with money in it; the director confirms anything creative.
- Keep a record. Know who accepted what and when, so you can trace a mistake back to where it entered.
- Limit what it can see. Not every tool needs the budget, crew rates or contracts. Give access on the same need-to-know basis you would give a person.
- Keep the creative calls human. Casting, performance, the look of the film and the edit belong to the people making it.
What about scripts, data and actors' likenesses?
An unreleased script is confidential, and so are budgets, rates and contracts. Before you paste any of them into a tool, find out where the data is stored, who can see it and whether it is used to train models. If the answer is unclear, assume it is not private.
Actors' faces and voices need particular care. If storyboard or previs frames use a cast member's likeness, get their consent first, explain how the images will be used, and keep those frames inside the production. Treat a likeness the way you would treat a contract: it belongs to the person, not the project.
Should AI ever touch the final cut?
Our view is no. The edit is where a film's meaning is decided: what to keep, what to lose, how long to hold on a face. Review tools can help people work on the cut, for example by collecting timecoded notes from everyone in one place, but the choices in the cut belong to the director and editor.
Used this way, AI is a fast and occasionally wrong assistant that handles the paperwork so the people on the production spend more of their time on the work only they can do.
Frequently asked questions
Can AI do a script breakdown?
AI can do a useful first pass, suggesting cast, props, wardrobe, vehicles and effects for each scene. A first AD or producer still needs to check every tag, add what the script implies but does not say, and settle open questions with the director.
Is it safe to upload an unreleased script to an AI tool?
Only if you know where the data is stored, who can access it and whether it is used for training. Treat the script as confidential, and if a tool cannot answer those questions clearly, do not upload it.
Will AI replace assistant directors or line producers?
It can take over some clerical tasks, such as first-pass tagging and drafting, but scheduling, budgeting and running a set depend on judgment, relationships and accountability. Those stay with people.
