Generation is cheap per attempt and expensive per keeper. A pipeline that renders sixty frames to ship four has a real unit cost fifteen times the sticker price. Studios quoting the sticker price are either new or hoping you do not ask.
Tracking it changes behaviour. Once the ledger exists, the argument stops being about which model is best and starts being about which gate is failing most, because the gate that fails most is the one burning the budget. That is a fixable, boring, extremely profitable question.
How do you budget an AI video production?
Agree a credit ceiling per asset before a batch starts and log every run. The job halts and asks for a decision at the ceiling rather than quietly spending past it.
Is generative production actually cheaper?
For the right brief, substantially: no crew, no location, no reshoot. For a brief that needs a real face, a real place and a real product on the same day, often not. We say which one you have before quoting.