Budget for accepted footage and the work needed to finish it. Normalize the exact endpoint, request type, duration, resolution, audio, and export requirements before comparing generation prices.
Quick answer: price the finished deliverable
A quoted cost per generated second is one input to a video budget. The practical total also depends on how much footage you discard, whether the service charges for inputs or finishing, and which plan permits the required export or use. Compare the cost of accepted footage under the same delivery requirements.
This guide explains provider billing structures.
Count generated seconds separately from usable seconds
Consider an illustrative project needing four eight-second shots. If each shot takes three paid generations, you pay for 96 generated seconds to obtain 32 seconds for the edit, before any additional work. The example assumes every generation is billable and every chosen take is usable in full; it is arithmetic, not a prediction about any provider.
Track total generation spend divided by accepted seconds, then record preparation and finishing separately. If you keep only part of a take, use the duration actually retained. For projects with no accepted take yet, report spend and rejection reasons rather than presenting a misleading efficiency figure.
Normalize the product and request type first
Record the exact service, model identifier, access route, and request type beside each quote. An application subscription, its developer API, and a third-party model offered inside that application can have separate balances and billing rules.
Runway Dev’s pricing table illustrates the differences: some models charge for output alone, while selected hosted models also charge for input video or reference assets. Aleph 2.0 has a minimum generation charge. These rules cannot be reduced to one universal “Runway price.” Use the Runway pricing record to identify the scope of a stored figure, then check the live table.
Do not assume duration scales linearly
Luma Agents’ pricing lists separate Ray 3.2 generation totals for five- and ten-second clips. In its standard-dynamic-range generation table, ten seconds costs three times the five-second total at the same resolution. Video editing uses a separate table; single-keyframe extension bills as one five-second block.
The budgeting consequence is straightforward: price the actual operation rather than multiplying a short-clip quote by your final timeline length. Splitting a shot can also create editorial work, so a cheaper set of requests is not automatically a better production plan. See Luma’s pricing notes before comparing request types.
Include the version you must hand over
Put upscaling, reframing, sound, and delivery formats on separate budget lines when required. Runway’s Gen-4.5 web guide says ProRes and PNG-sequence output is restricted to specified plans and adds a per-second charge at generation time. That is a different proposition from downloading the base output.
Pika 2.5’s model page says its renders are silent. A soundtrack is therefore a separate workflow consideration for that model. Estimate editing, audio preparation, review, and export labor alongside service charges. Keep a cash-spend total and a labor estimate rather than pretending your time is free.
A subscription is economical only if the entitlement fits
Pika’s current plan page says plan credits do not roll over, while purchased top-ups expire after 365 days. It also says Create top-ups cannot be used with the API. The same page lists no commercial license for Free or Starter, and includes one for Creator and Fancy.
Those details matter more than dividing a plan price by its largest advertised output count. Check your expected monthly usage, whether unused value expires, the billing commitment, and the entitlement required for the intended project. Do not assume that paying for a plan includes every permission. Compare the Pika billing record and commercial-use notes with the current product terms.
Distinguish a failed request from an unwanted result
Pika’s image-to-video API pricing says only successful runs are charged. Luma’s failure policy distinguishes synchronous errors and several refunded asynchronous failures from budget-exhausted failures, which can incur a partial charge.
A technically successful clip that does not meet your brief is a different issue. Do not assume a failure refund applies to a creative rejection. Keep request status, invoice charge, and editorial acceptance as separate fields. If a charge is unclear, reconcile the provider’s request record before treating a retry as free.
Availability can change the real cost
A low historical rate is not a current purchasing option. OpenAI’s deprecation record lists Sora 2 and the Videos API as shut down September 24, 2026. Google’s video guide schedules Gemini API Veo 3.1 Preview shutdown for October 22, 2026, while pointing to other access routes and models.
Before committing production work, allow for migration and repeat evaluation if the endpoint is nearing retirement. Do not carry the old price, prompt behavior, or quota assumptions into a replacement without checking them.
A compact budget checklist
- Define the accepted deliverable: shot count, retained duration, ratio, resolution, sound, and file format.
- List the exact model, service, request type, and dated billing source.
- Set a limited evaluation allowance and record every billable attempt.
- Add reference preparation, editing, finishing, and required plan commitments.
- Reconcile actual charges, unused credits, taxes, and currency conversion where applicable.
- Review accepted footage and total effort before expanding the budget.
Official sources were checked October 21, 2026. Use the documented billing structure to build a budget for your own deliverable.
Before you choose a tool
- Define the output you need before comparing plans.
- Save the inputs, settings, date, and rejected results.
- Check current provider terms and export restrictions.
- Include correction time in your comparison.
Sources
Official sources reviewed 2026-10-21.
