The listed price of a generation request is only one part of production cost. A more useful measure is spending per usable asset: total generation spending divided by the number of outputs that pass your review. This makes retries and rejected results visible instead of hiding them behind a low per-request price.
Define “usable” first
Write the acceptance criteria before running the test. An image might need correct product details, the right composition, and no visible defects at delivery size. A video might need a minimum usable segment with stable identity and acceptable motion. Keep those criteria consistent across candidates.
If one output is accepted only after manual cleanup, record the cleanup rather than treating it as equivalent to a ready-to-use result. You can track provider spending and human effort separately while still considering both in the final decision.
Use a simple calculation
Start with Nova Creative's current plans and generation routes and the selected provider's charges. The BYOK planning guide explains how to separate subscription costs from provider usage.
Generation cost per usable asset = total generation charges / accepted assets.
For an illustrative example, imagine twenty attempts cost a total of $8 and four images pass review. The generation cost per usable image is $2. These are invented numbers for the calculation, not Nova Creative or provider prices. If no output passes, report zero accepted assets and the spending incurred; dividing by zero does not produce a meaningful unit cost.
Keep a small production ledger
- Record each model or workflow candidate.
- Count all attempts, including revisions and rejected outputs.
- Use actual billed spending when available.
- Count assets that pass the same acceptance criteria.
- Record review and cleanup time separately.
Also record currency and the time period. Do not combine charges in different currencies without a stated conversion method. For subscriptions, make the allocation method explicit: project-specific usage, a share of the month, or another consistent basis.
Compare like with like
A short draft clip and a finished campaign video are different outputs. If you are comparing video generators, record usable duration and the finishing work still required. A cheaper clip may become more expensive if it needs repeated trimming, replacement shots, or extensive repair.
Similarly, compare equivalent image sizes and production goals. A thumbnail concept and a detailed product image should not share an acceptance standard unless they truly serve the same job.
Use the result to change the workflow
If spending is high because the brief is unclear, refine the shared input before switching providers. If one material repeatedly fails, route those items to a different process. If a model creates attractive but inaccurate products, count those outputs as rejected.
In Nova Creative, keep the graph small enough to understand the planned requests, and review the selected generation route before expanding branches. The goal of this metric is a better production decision: which workflow reliably produces the assets you can use, at a cost and review effort you can sustain.