Why brands are hiring AI film studios instead of production companies

Two numbers from the same industry, eighteen months apart. A thirty-second spot for the prediction market Kalshi ran during the 2025 NBA Finals. One person made it in three days for around two thousand dollars. The Toys"R"Us origin film, one of the first commercials made with Sora, took about twelve people roughly three months — which NBC News noted was roughly the same as any other job.

Made in

3 days

Crew

One

Generations

300–400

Both are true. Neither is the headline. The useful question is not whether AI makes advertising cheaper, because sometimes it does not. It is what a brand is actually buying when it hires a studio like this one, and which part of the old arrangement quietly disappeared.

The thing that changed is revision, not price

On a conventional shoot, the budget is exposed on one day. Crew, location, talent, equipment and weather all arrive at once, and everything that happens afterwards is selection from what you got. A change of mind after wrap means a reshoot and a new line in the budget, which is why most changes of mind simply do not happen.

Generation moves that cost. A shot can be rebuilt forty times for the price of rendering it forty times. The film gets decided in the edit rather than on the day, and the client who wants the hero's jacket in navy on Thursday is no longer asking for something unreasonable.

That single change explains most of the others. Scale follows from it: once a shot can be regenerated, forty cutdowns, twelve aspect ratios and six languages with matched lip sync cost a fraction of the first version rather than a multiple of it. So does the thing nobody advertises — the number of attempts. The Kalshi spot took three to four hundred generations to yield fifteen usable clips. Coca-Cola's 2025 holiday films ran to more than seventy thousand generated clips across two films, with around a hundred people involved.

Read those two together and the shape of the work becomes clear. Generating a shot is easy. Knowing which of four hundred takes belongs in the film is the job, and it is the same judgement a director has always been paid for.

Where the saving is real

The clearest published figure comes from the American agency Copacino Fujikado, which told Digiday it delivered a campaign requiring twenty-six models across several locations for twenty-six per cent of the live-shoot cost. That is a genuine saving, and it is worth noticing why. The brief was logistics-heavy. Most of what a conventional production would have spent went on moving people and equipment to places, and that is precisely the part that disappears.

Apply the same logic in reverse and you can predict where it will not save you anything. A single narrative film with one hero character and a story to tell has almost no logistics to remove. Svedka's Super Bowl LX spot — the first primarily AI-generated national Super Bowl ad — took about four months, most of it spent rebuilding one character so she stayed herself from shot to shot. Sazerac's own marketing lead said plainly it was not about saving time or money.

So the honest version of the pitch is narrower than the one you have been hearing. AI production saves money where the alternative involved moving a lot of people to a lot of places, or where you need many versions of the same idea. It saves very little on a single beautiful film, and anyone who tells you otherwise has not shipped one.

The work that actually ships is hybrid

This is the part that gets lost. The best-documented AI production of 2026 is not a prompt. The studio Promise reconstructed ten real crime scenes and delivered more than five hundred AI-generated shots for the second season of Harlan Coben's Final Twist, which premiered on CBS on 5 October 2026. The pipeline was generative AI plus 3D, Gaussian splats and Unreal Engine, with the host filmed conventionally on a volume stage. Netflix, in its Q2 2026 shareholder letter, put generative AI in roughly three hundred titles, concentrated in post-production, and said some shots would have been left out entirely without it.

Set that against what has not shipped. Lionsgate and Runway announced a custom model trained on a catalogue of more than twenty thousand titles two years ago. Nothing has been released. A source close to it told TheWrap the problem was data scale — "the Lionsgate catalog is too small to create a model".

“Knowing which of four hundred takes belongs in the film is the job, and it is the same judgement a director has always been paid for.”

The gap between the announcement and the delivery is now measurable, and it points the same way every time. The work that reaches an audience is made by people who treated generation as one tool in a pipeline. As Dave Clark, who runs Promise, put it: you cannot prompt your way to the ends.

The evidence that cuts the other way

We would rather you heard this from us. In May 2026 Ipsos and Syracuse University tested twenty ads across ten brands, pairing each brand's human-made thirty-second spot with a fully AI-generated version built from the same brief, in front of three thousand consumers. The human versions scored fourteen per cent higher on short-term sales impact and seventeen per cent higher on brand equity.

The interesting finding is not that gap. It is that only a quarter of viewers were even somewhat confident they were watching something AI-generated. Audiences largely could not tell, and bought less anyway. Where AI did best was product-driven, direct briefs. Where it did worst was storytelling, emotion and point of view.

That result is a description of the job, not an argument against it. It says the machine is good at the part that was always mechanical and weak at the part that was always authorship. It is also the reason this studio sells direction rather than throughput.

What to ask before you brief anyone

If you are choosing between a production company and a studio like this one, the brief decides more than the supplier does. Three questions get you most of the way.

• How many finished versions do you need? One film, one cut, one market is the weakest case for this. Forty versions in nine languages is the strongest.
• Is there a thing that has to stay exactly itself? A recognisable character, a real product in a real hand, a face the audience already knows. That is where the hours go, and where a quote should go up rather than down.
• What happens when you change your mind in week three? On a shoot the answer is a reshoot. Here it should be a new render. If a studio cannot tell you plainly which of the two it is, it has not planned the job properly — Knucklehead's Tim Katz puts it simply: changing a shot late is really hard, so it has to be designed in from the start, not bolted on.

The budget did not shrink. It moved out of the shoot day and into direction, iteration and finishing. What you are buying is the ability to keep deciding — and the taste to know when to stop.