Every brief gets the same question early: which shots should be filmed, and which generated? Get it wrong one way and you spend a shoot day on something a model could do. Get it wrong the other way and your audience stops believing the film.
We shoot what should be shot, and generate what should be generated.
The rule
- Shoot it when the shot depends on proof: a real face, place or product the viewer needs to trust. Testimonials, founders, food, a product doing what it says.
- Generate it when the shot is impossible, expensive or variant-heavy: a world that doesn't exist, a scale no budget reaches, or twenty versions of one idea.
- Go hybrid when the shot needs both: a real hero inside a world the camera can't reach.
The skill is applying it shot by shot, not film by film. Most briefs aren't all one thing.
Why proof means a camera
A viewer reads a filmed face differently from a generated one, even when they can't say why. When the point of a shot is "this person really said this" or "this product really looks like this", generating it undercuts the point. A generated product shot that looks better than the real thing is a misleading-conduct problem waiting to happen.
So for proof, we shoot it for real. Live-action is one of our lines for that reason, not a leftover from before AI.
Why impossible means a model
Generative AI earns its place on three kinds of shot.
- Impossible: a city that doesn't exist, a product not yet made, a reshoot that can't happen.
- Expensive: a set build, an overseas location, a crowd or an aerial, for a few seconds of screen time.
- Variant-heavy: one concept re-cut for many markets, formats or audiences, where shooting each version would cost ten shoot days.
AI-assisted and art-directed by people, these shots can be finished to the same standard as anything we film.
Three examples
These are typical briefs, not client case studies.
A financial services brand wants its founder telling the story of why the business started. That's proof from start to finish: a real person, their own voice, their own office. We shoot it.
A drinks launch wants its can on a floating island above a neon city, in five formats. Only the can needs to be real, so we generate the scene, match the can to the approved artwork, and have the client sign off the product in every shot.
A tourism campaign needs a real family on a real beach, with a sunrise in every market and a shoot window of one grey week. The family is proof, so we shoot them. The sky, extra coastline and market versions get built. That's hybrid.
What the rule protects
The rule is about the film working, not cost first. It usually lands on the leaner answer too, because each shot gets the method that suits it.
It also protects the work over time. Models change monthly. OpenAI gave notice on 24 March 2026, then removed Sora 2 and its Videos API from the API on 24 September 2026, stranding any workflow built on that one model. A studio that can shoot, and keeps generated work able to move between models, can keep a campaign running when a tool disappears.
A Hybrid Day, one shoot day plus AI set extension on up to 10 shots, is $12,500 ex GST: the shoot covers the proof and the model covers the rest.
Who decides
We recommend a method for each shot at treatment stage, with the reason beside it, and you make the call. If a real face would earn more trust, we'll say so, even when the brief asked for AI.
We do both, so we have no reason to push you towards either.
