The new temptation goes like this. Generation is cheap, so why sit with an idea when you can make shots right now and see what sticks? A mountain of shots later, you have a very expensive mood board and still nothing to say.
AI made production faster and cheaper. It did nothing for the idea. A weak idea rendered beautifully stays weak. It arrives sooner, with a nicer grade, and audiences scroll past polished nonsense with remarkable efficiency.
Cheap shots make the idea more important
When every brand can generate a slick shot, slickness stops being a reason to watch. What separates one ad from another moves back to where it always was: what the ad is saying, and whether anyone cares. Cheap production also removed a favourite excuse. Nobody gets to say the budget killed the idea any more.
There is a second trap. Generating shots feels like progress. The timeline fills up, the team gets excited, and somebody starts choosing music. By the time anyone asks what the film is about, too many hours have gone in for anyone to admit it is about nothing. The fix is boring: no generation until the idea has passed a few small tests on paper.
Test the idea with one line, one image, one feeling
Before a single shot, the idea should pass three small tests. They cost less than one round of generation and they save most of the rounds after it.
One line. Write the idea as one sentence a stranger would understand. Leave the tagline and the brand promise out of it and write what happens. Say you sell a cooking app. “A grandmother teaches her grandson to cook over a video call, and he burns everything” is an idea. “Celebrating connection across generations” is a filing category. If everyone on the team writes a different sentence, you have several ideas and no decision yet.
One image. Describe the single frame people would remember if they forgot everything else. If you cannot picture it, the film will not have it either. If you can, sketch it or make one rough frame. Keep it deliberately rough, so the conversation stays on the idea.
One feeling. Name what the viewer should feel at the end. Amused, reassured, hungry, a bit proud. Choose one. Ads that chase four feelings tend to land none of them.
Show it to someone who owes you nothing
Take the line, the frame and the feeling to people outside the team. Read them the line and show them the frame. Ask what the ad is about and how it made them feel. Skip “do you like it”, because people are kind and kindness is useless data. If their answer matches yours, you have something. If they describe a different ad, the problem sits upstream of production, and no model will fix it.
Kill weak ideas at this stage, while it is cheap and nobody has fallen for a shot yet. Dropping a sentence is painless. Dropping a finished sequence the client has already seen is a very long call.
Then let AI do what it does well
Once the idea survives, generation becomes useful. You can explore more visual directions for the same idea than a conventional budget would allow, and build shots no physical shoot could reach. That is where AI earns its place: serving an idea that already works.
Keep the line pinned to the top of the brief while you generate. Every shot has to earn its place by serving that sentence. When a stunning shot does nothing for the line, save it for another film and move on.
People direct, and AI builds the shots. That order is how AI production should run, and it starts with an idea someone can say out loud in one breath. If yours needs a paragraph and a mood film to explain, go back to the line.