The cheapest generation can cost more per deliverable when every approved take needs multiple rerenders. Compare approaches on credit spend divided by accepted results for the same reference shot. Make the acceptance rule before seeing which input setup looks more promising.
Build the test input
Use one consented hero asset and a clear requirement: the subject turns halfway while retaining a visible collar pin and ending inside an agreed crop. Compare a single visual reference with a richer reference pack only where both are supported. Set a small, equal attempt limit for each condition.
Run the proposed comparison
Use displayed credit quotes from the actual studio and store actual spend when the experiment finishes. Keep duration, crop, and audio fixed. For every response, record approved, rejected, or billed failure. Keep refunded attempts distinct because a retry can still consume time without consuming final credits.
Decide what passes
For illustration, 500 spent credits divided by two approved clips gives 250 credits per approved clip. A condition with no approved takes has no finite observed cost per accepted clip at that budget. Inspect whether any acceptance differences reflect one reviewer relaxing the collar rule.
Act on the outcome
Prefer the setting that meets the deliverable within your time and spend limits. Where the credit costs are similar, compare the total waiting and review time. Repeat on another product if you plan to extrapolate beyond one source asset, and retain the rejection reason for each take.

