A generated minute and an approved minute are not the same product. A project is delivered when its shots are usable, fit the brief, pass review, and occupy the required duration. A tool's cost-per-generated-second is only one input. The more useful business unit is credits per usable second.
Start with a delivery requirement
Imagine a hypothetical three-minute brand film. The editor must deliver 180 seconds of approved material. The shot plan uses six-second clips, so the minimum count of accepted shots is ceil(180 / 6) = 30. That is the delivery requirement; it says nothing yet about the number of attempts needed.
We will assume 2.5 attempts per accepted shot as a planning hypothesis. It is not a published performance benchmark for any AI model. At a purely illustrative rate of 5 credits per generated second, the working estimate is:
| Quantity | Calculation | Example result |
|---|---|---|
| Required approved shots | 180 ÷ 6 | 30 |
| Planned generations | 30 × 2.5 | 75 |
| Total generated seconds | 75 × 6 | 450 s |
| Credits consumed | 450 × 5 | 2,250 |
| Credits per usable second | 2,250 ÷ 180 | 12.5 |
The nominal rate in this example is five credits per second. The cost of the deliverable is 12.5 credits per accepted second. The gap is the planning allowance for rejected or revised generations, not necessarily wasted spending: rejected outputs may provide useful creative evidence.
Add a range, not a single confident number
A single estimate can hide uncertainty. Estimate three retry scenarios, then choose a reserve you can afford:
| Attempts per accepted shot | Calculated credits | Meaning |
|---|---|---|
| 1.5 | 1,350 | Optimistic: unusually efficient approvals |
| 2.5 | 2,250 | Working plan: a moderate retry allowance |
| 4.0 | 3,600 | Stress case: difficult art direction |
These rows share the same six-second duration and five-credit example input. If actual generation lengths are rounded into fixed duration tiers, replace the six seconds with the billed duration, not the desired trim length. A 4.5-second approved segment may still incur a bill for a longer source clip.
Include work outside the model
Credit cost is not total project cost. A credible estimate also reserves time or money for script changes, image preparation, voice licensing, editing, captions, sound design, quality checks, and revision rounds. Keep these as separate line items so a producer can see what is actually being charged.
Do not convert the credits to dollars unless you know how the chosen plan, purchased credits, and remaining balance work. A credit on a monthly allowance may not have the same effective cost as a top-up credit. When those facts are unavailable, publish the credit forecast and leave money as an editable input.
Use a release gate
Before purchasing a larger plan, produce a small set of representative shots. Record the prompt version, attempts, accepted seconds, and reasons for rejection. Then replace the hypothetical 2.5 with your own observed average—labeled with the number of test shots. If the test has only two or three shots, keep a wide uncertainty range.
Decision rule: approve the budget when your low, working, and stress scenarios are visible and you can explain where the retry assumption came from. Our project calculator lets you replace every example input with your own.
Check the source and keep your inputs
Our examples show a decision-making method, not a live price quotation. Before purchasing credits, verify the provider’s billing method and current plan rules in its official credit documentation or the equivalent official page for the tool you use. Keep your calculations reproducible with our shot-by-shot CSV worksheet. See the methods and sources page for limitations and a full cost ledger.