What AI Still Cannot Do In An Ad Film
Generative video closed the gap on look far faster than the industry expected. A single beautiful shot is now cheap. What has not closed is everything that makes thirty seconds hold together as one piece of film.
An ad film is not a sequence of good shots. It is a controlled argument with timing, and the parts that are still hard are the parts that were always hard for a reason.
The Three Things An Ad Film Still Needs A Crew For
Continuity across cuts, first. A face, a garment and a room have to survive a change of angle, and generated shots drift in ways that read as wrong even to a viewer who could not say why.
Performance, second. Not expression, which models render well, but the small delay before a reply, the reaction that arrives slightly late. Comedy in particular runs almost entirely on timing that has to be felt rather than specified.
The Third Is The Expensive One
Direction. Someone deciding this take and not that one, for reasons they can articulate to a client afterwards. Generation produces options, and options are not decisions.
The bottleneck moved. It used to be making the shot. Now it is knowing which of two hundred shots is the one, and that has not got any cheaper.
Where This Leaves Production
The economics have split rather than collapsed. Anything atmospheric, textural or short is now realistically generated. Anything that depends on a person being believable over several cuts still wants a set.
What that means in practice is a split brief rather than a smaller one. Atmosphere and texture get generated, performance gets shot, and somebody has to hold the two together so the seam does not show. Read alongside the production cost picture, the pattern is consistent: cost has fallen fastest exactly where judgement matters least.