Budget for 10 Usable AI Videos, Including Rejections

A generation price tells you the cost of an attempt. A campaign needs usable outputs. Connecting those two units requires an explicit assumption about rejection, as well as the work that happens after a clip is accepted.

Open the ten-clip production example. The default inputs reproduce the base case below. Choose a current billing rate from your provider before using the result as a project estimate.

Define the deliverable before choosing the rate

Assume the campaign requires ten deliverables with one accepted variant each. Every generated clip lasts eight seconds. The hypothetical provider rate is $0.10 per generated second, so an attempt costs $0.80. There are no fixed production charges in this base case.

If 25% of attempts are accepted, the expected number of generations is ten divided by 0.25, or 40. Generation spending is therefore $32, not the $8 obtained by pricing only ten successful attempts.

Each accepted clip also requires three minutes of editing at $20 per hour. That is $1 per accepted clip, or $10 for the campaign. The total expected budget is $42, equivalent to $4.20 per accepted clip.

const acceptedClips = 10;
const acceptance = 0.25;
const generatedClips = acceptedClips / acceptance;
const generation = generatedClips * 8 * 0.10;
const editing = acceptedClips * 3 / 60 * 20;
console.log(generatedClips, generation + editing); // 40 42

Acceptance can dominate a small rate difference

Hold the clip duration, rate, and editing assumption constant, then change only the accepted share.

Accepted share	Expected generations	Generation cost	Total with editing
50%	20	$16	$26
25%	40	$32	$42
10%	100	$80	$90

The table does not claim any particular model achieves these rates. It shows why a cheaper generator can fail to reduce production cost if more outputs are rejected. A proper comparison needs the same creative brief and the same acceptance criteria.

Also distinguish an accepted variant from a generation request. If every deliverable needs three accepted variants, the target becomes 30 accepted clips rather than ten. Under the base assumptions, both generation and per-accepted-clip editing costs triple, while a separate fixed project charge would not.

Expected cost is not a spending cap

The equation assumes attempts behave independently and have a constant acceptance probability. Actual projects can violate both assumptions. A difficult shot may fail repeatedly, or a revised prompt may improve later attempts. The calculator does not estimate those effects automatically.

Forty generations is an average in the model, not a promise that the fortieth attempt completes the campaign. A production stop rule, contingency allowance, or decision to simplify a shot belongs in the project plan. Do not label the expected total as a worst-case budget.

At zero acceptance, no finite expected number of generations produces the requested accepted clips. Showing a zero-dollar project in that situation would be an arithmetic and product-design error, so the tool reports that no finite estimate exists.

Put the remaining production work in the right place

The editing input is per accepted clip. Review of rejected clips is not automatically included. Either incorporate that work into your generation-side unit cost or add a separately calculated production amount without counting it twice.

Audio, upscaling, storage, resolution-specific pricing, minimum billed duration, and other charges may use different units. The media pricing explorer can help identify the source rate and unit, but a single dollars-per-second input should not silently absorb incompatible pricing rules.

If a provider bills whole requests rather than seconds, calculate the effective cost of the exact request configuration you are modeling. Changing duration or resolution can invalidate that conversion. Keep the original billing unit next to the assumption in your estimate.

Sources and method

This article is an original expected-cost scenario. All acceptance rates, labor assumptions, and generation rates are illustrative. It uses accepted targets divided by acceptance probability, then adds generation, editing, and fixed costs. The base case and sensitivity rows are checked against the same calculation engine used by the tool.