A brand asks for a campaign visual. Twenty minutes later there is an image on screen that would have taken a week and a shoot five years ago. Everyone in the room agrees it is remarkable. Then someone asks for it in portrait, with the product two shades warmer, the logo clear of the shoulder, and eleven more like it. That is where the actual work starts.
The four things nobody generates
Almost every AI campaign that stalls, stalls in the same four places.
- Consistency. One frame is a result. Forty frames that look like the same campaign is a system, and systems have to be designed.
- Exactness. The product has to be the product. Not a convincing approximation of it with the wrong number of stitches.
- Brand fit. Generated material has an accent. Removing it and replacing it with the client's accent is art direction, not prompting.
- Delivery. A campaign is not an image. It is a matrix of formats, resolutions, safe areas, languages and versions.
One good frame is luck. Two hundred consistent ones is a pipeline.
What the 80% actually consists of
Selection, first — knowing which of four hundred outputs is worth building on, which is a skill that predates the tools by decades. Then the repair work: geometry that will not hold up, materials that read wrong, type that has to be set rather than hallucinated. Then the parts that get built in CGI because they must be accurate. Then compositing, retouching, grading, and the unglamorous business of versioning and delivering files that a media agency can actually run.
None of this is a criticism of the tools. It is a description of what production has always been. The generation step simply moved a large, expensive part of the process — exploration — from weeks to hours. That is a genuine shift. It is also not the same thing as finishing a campaign.
Why this matters commercially
Teams that budget for the 20% and assume the rest is free tend to spend the difference twice: once on the generation, and again on the agency that has to make it usable. Teams that plan the whole pipeline get the actual benefit, which is real and large — more directions explored, fewer compromises locked in early, less money spent proving out ideas that were never going to work.
The value was never in access to the tool. It is in knowing what to make of what it gives you.