Project planning · 6 minute read

How to budget an AI video project

A project budget is easier to reason about when you separate finished runtime from footage you expect to generate. Scenes, alternate takes, and a contingency allowance each affect the number of API calls.

1. Set a finished runtime and scene length

Decide how long the delivered video should be, then choose a target generation length for each scene. Divide the finished runtime by scene length and round up: a partial final scene still has to be generated. A 90-second edit made from 6-second scenes is 15 scenes. A 91-second edit at the same scene length needs 16.

2. Decide how many attempts to budget

One attempt means one generation request for each planned scene. If you want two candidate takes per scene, multiply scene count by two. The number is a planning assumption: some scenes may be accepted on the first try, while others need more iterations. Record this separately from the contingency percentage so the same allowance is not counted twice.

3. Price one attempt at the exact endpoint

Check the provider’s billing unit and match resolution, audio, mode, and duration. When billing is per second, a six-second generation at an illustrative rate of $0.10 per second is $0.60 per attempt. Using 15 scenes and two attempts gives 30 attempts, or $18 before any buffer. This rate is hypothetical for explaining the arithmetic; it is not a current provider quote.

4. Add a clearly defined buffer

If you add an illustrative 20% contingency to that $18 base, the planning estimate becomes $21.60: $18 × 1.20. The buffer can cover unpredictable iterations and small production changes, but it does not replace provider-specific retry, cancellation, or failed-request rules.

5. Test a scene before scaling up

Generate a small sample using the settings you intend to use. Confirm how the endpoint bills actual duration, whether audio adds cost, how minimums or rounding work, and what happens when a request fails. Compare those results with the provider’s current documentation, then update project assumptions before generating all scenes.

Keep the estimate in scope

Generation cost is only one part of a video budget. Add separately billed input processing, storage, upscaling, editing, music, voice, taxes, or delivery charges from their own sources. If a project uses several models or API hosts, estimate each group of scenes with its own rate instead of assuming one rate applies to the whole edit.

Quick planning formula: estimated scenes × attempts per scene × cost per attempt, then multiply by 1 + buffer percentage ÷ 100. When rates, minimum charges, or endpoints differ between scenes, calculate each group separately and add the subtotals.

Use the Project Calculator → · Read how billing units are handled →