Related: See our AI-driven work · Start a project
I direct commercials, and I build AI systems. That combination is rarer than it should be, and it has made me less impressed by AI in production, not more — because I know what the tools are doing underneath, and I know how far that is from what a shoot demands.
Almost everything written about AI and filmmaking is a prediction. This is an inventory.
What it genuinely absorbed
Pre-visualisation and boards. This is the clear win, and it isn't close. Generating look options in minutes instead of days changes what happens in a client conversation — you stop describing an idea and start pointing at one. Ambiguity is the most expensive thing in prep, and this removes a lot of it before anyone is paying for a crew.
Location scouting, first pass. Narrowing a hundred candidates to a shortlist worth physically visiting. Not a replacement for the recce — a replacement for the wasted half of the recce.
Transcription, logging, and rough assembly. Searchable transcripts across hours of footage, on the day. On documentary work this is transformative; finding the moment used to be the job.
Versioning and deliverable wrangling. The unglamorous end — aspect ratios, cutdowns, localisation passes. Genuinely faster, and genuinely lower-value work to have been doing by hand.
Paperwork. Schedules, call sheets, breakdowns, first-draft treatments. The administrative substrate of production, compressed.
Notice what these have in common. They are all upstream or downstream of the shoot. None of them happen while a camera is running.
What broke
Continuity. Generative tools do not hold a thing consistent across shots. A face, a garment, a product — the exact things a brand cares about most are the things these models are worst at keeping stable.
The product. A client's product has a defined shape, finish and colour, and it is the one element that cannot be approximately right. Every generative pass I have seen still requires the real thing, shot properly.
Anything requiring a decision under time pressure. The tools have no judgement about what matters. On set, the entire job is deciding — fast, with incomplete information, while twenty people wait. There is nothing in the current stack that does that.
What stayed human, and will
Directing is mostly a social act. It is getting a performance out of someone who is nervous, or bored, or being asked to do the same thing for the ninth time and make it look like the first. It is reading a room of clients and knowing which note is the real note. It is telling a brand their idea won't work, and being believed.
None of that is a generation problem. It is a trust problem, and trust hasn't been automated.
The other thing that stayed human is taste — knowing which of the forty generated options is the good one. AI has made options cheap, which has quietly made judgement more valuable, not less. When everyone can produce a hundred directions in an afternoon, the constraint moves to whoever can tell which one is right.
The honest summary
AI took work off my plate. All of it was work I would have been happy to lose — logging, versioning, paperwork, the first ninety percent of a scout.
It has not touched the day. The shoot is still a room full of people solving problems in real time, and the director's job is still to be the person who decides.
I would be wary of anyone selling you the opposite. In my experience, confidence about AI in production correlates almost perfectly with distance from an actual set.
Read next: How to Compare Three Production Bids That Look Identical · What Makes a Great Creative Brief
Static Free Films is a director-led, minority-owned production company in Los Angeles and London. Brand work includes Nestlé, Microsoft, HP, adidas, Samsung, HelloFresh, WWF and GE.
Working out where AI genuinely fits in your production pipeline? Start a project and I'll tell you what I've seen work and what I've seen waste money.