If a shot needs to convince someone that a product really does something, I lean towards live action. The more important that detail is to the viewer's decision, the less I want to leave its accuracy unresolved until the end.
AI can change the settings and images available to a project. I still decide how to make each shot by looking at its job in the film.
What does the product actually do?
A button press, a hand picking up an object or a lid opening may look simple. Each involves contact, force and the movement of separate parts.
If that action demonstrates the selling point, I would establish it clearly first, then decide whether to extend the surrounding scene. The most persuasive part of the film has a dependable basis, and the creative work has somewhere to begin.
For packaging, ports and close product views, I also favour approved photography or accurate assets. When someone may compare the video with the item they receive, a close approximation is not enough.
Where I would explore with AI
Atmosphere, unusual spatial scale and conceptual transitions can be useful places to test generated imagery. I want to know why the world changes, and what that change has to do with the product.
Consider a hypothetical luggage film moving from a bedroom into an unfamiliar city. I would favour filming the handle extending, the wheels crossing a threshold and a person lifting the case. The changing city outside the door could be explored separately with AI. This is a proposed division of work, not a completed project.
A generated environment still has to belong with the subject. Camera height, light direction and contact with the ground matter. If those disagree, a floating suitcase may attract attention before the city does.
Test the difficult shot
An easy atmospheric shot tells me little about a difficult product interaction elsewhere in the script. If a sequence depends on a large turn, close contact or complicated occlusion, that is the test I would want to see.
I would ask two questions: is this result usable, and can it be revised as the brief requires?
A request to hold a hand still for longer should not be judged only by the hand. If the next output changes the packaging and background as well, the method needs another look. One attractive result is a starting point, not proof that the whole film should be made the same way.
| Shot's purpose | Starting approach | What the test must establish |
|---|---|---|
| Explain real operation | Live action or accurate assets | Product and action are clear |
| Establish a conceptual setting | AI visual development | The setting serves the idea and fits the subject |
| Connect real and imagined elements | Layered compositing and editing | Perspective, light and the transition work together |
| Present the final product and text | Verifiable product imagery, independently handled text | Each delivery version remains accurate and readable |
There is no fixed percentage of AI footage to aim for. A project may need one generated shot or many. I would settle the parts that depend on accuracy first, then decide how far the visual idea can go.
I am the founder of ONCE. Our AI and live-action production page explains the team's work. The choice for an individual film still starts with its product, script and tests.