What generative video changed, and what it did not

The question people ask about generative video is whether it replaces production. It is the wrong question, and it produces bad decisions in both directions — teams that shoot nothing and teams that refuse to touch the tools at all.

A more useful framing: generative tools did not make ideas cheaper. They made variants cheaper. Those are different things, and confusing them is how briefs end up with forty assets and nothing worth watching.

The shot list is still the hard part

Generating a shot takes minutes. Knowing which shot to generate takes the same work it always did — a position, an audience, a reason for the piece to exist. None of that moved.

What did move is the cost of being wrong about a shot. Reshooting used to mean re-assembling a crew, a location and a schedule. Now a shot that is not working can be discarded and rebuilt in the same session. That changes how early you can afford to be specific, and it makes the storyboard the most valuable document in the project rather than the least.

Where it genuinely helps

• Concept films made before a budget exists, so a client can see the idea instead of reading it
• Locations and seasons you cannot shoot — a monsoon rooftop in a dry month, a coastline nobody is flying to
• Product and styling variants, where the set-up is identical and only the subject changes
• Vertical recuts, where the format was decided after the shoot and the coverage does not exist

Where it does not

Continuity across a long cut is still difficult: a face, a garment or a room will drift between shots unless the piece is built to tolerate it. Real products are harder still, because a generated approximation of a real object is an approximation of someone's trade dress, and that is a legal problem rather than an aesthetic one.

And anything that depends on a real person, a real place or a real moment is not a candidate at all. A founder talking about their own company cannot be generated. Neither can a room your customers will recognise.

The new failure mode

The old constraint on content volume was cost. That constraint is largely gone, and what replaced it is worse: volume with nothing behind it. A feed does not reward output. It rewards the first second, and the first second is an editing decision, not a generation one.

Cheap variants are only useful once you have something worth varying.

How we treat it

As a production method, not a category of work. The short-form films on this site were produced this way — talent, set, product and grade generated shot by shot — and they are labelled as capability demonstrations rather than as client campaigns, because that is what they are. Where a film was made for a client, the client is named.

The useful test has not changed. Would you watch it if it were not yours? Generation makes it faster to find out, and slightly more tempting to skip asking.