What generative production actually changes, and what it does not
The cost of making an asset fell by an order of magnitude. Almost nothing else in the pipeline did, which is why most teams are not feeling the benefit.
iStudios, Creative practice
Generative tools collapsed the marginal cost of producing a creative asset. That is real and it is large. The mistake is assuming the marginal cost of producing an asset was what constrained you.
In most organisations it was not. The constraints were legal review, brand approval, trafficking capacity, rights clearance and a measurement setup that cannot resolve more than a handful of creatives at once. Speed up production without touching any of those and you get a larger queue, not more output.
A tenfold increase in asset supply against an unchanged approval pipeline produces a bottleneck with better graphics.
Where the economics genuinely change
- The long tail of formats. Every platform wants its own ratio, length and safe area. This work was always expensive and never creative, and it is the clearest win.
- Localisation. Twelve markets used to mean twelve production cycles. It now means one master and a versioning system, provided the brand rules are explicit enough to be encoded.
- Iteration inside a flight. Refreshing creative mid-campaign used to be a planning decision. It can now be an operational one.
- Concept exploration. Seeing twenty directions before committing is cheap, which changes what gets considered rather than what gets shipped.
Where it does not
The idea. The judgement about which of the twenty directions is worth making. The relationship between a brand and the person who trusts it. Generative systems are excellent at variation and poor at knowing which variation matters, and the gap between those two is where the value in creative work has always been.
The prerequisite nobody wants to fund
All of this assumes your brand rules are written down precisely enough to be encoded as constraints. Most are not. They exist as a PDF of principles and a designer who knows what is acceptable, which works fine at ten assets a month and fails completely at four hundred.
Writing that rulebook is the unglamorous prerequisite for the whole thing, and it is the project that gets cut first.