As checked 2026-10-25: A provider can sell AI video by generated second, tokenized video output, job credit, clip, or paid plan. This is a dated economics guide, not a permanent price guarantee or a personal hands-on benchmark.
Scope: global English reference; USD list prices where shown; local eligibility, currency, taxes, negotiated terms and model release dates require separate confirmation. Dynamic source-page prices have a check date, not a verified initial effective date.
Video rate cards hide several meters
A provider can sell AI video by generated second, tokenized video output, job credit, clip, or paid plan. Gemini Omni Flash documentation illustrates this distinction: the page publishes a video-output token rate and gives a 720p token-to-second conversion for its stated standard mode. fal’s model API listings instead show several video routes priced per generated second. A token tariff and a per-second price are not contradictory, but require a supported model-specific conversion and resolution assumptions.
From scene list to generation budget
A 12-minute final video does not necessarily require only 720 seconds of generated footage. Define final seconds, attempted seconds, accepted-seconds ratio, transition overlap, upscales and audio separately. If a storyboard needs 60 eight-second scenes, planned first-pass generation is 480 seconds; with 50% of scenes needing one additional take, the scenario reaches 720 generated seconds before any rendering overhead. This is scenario arithmetic only; it is not a measured rejection rate.
Resolution and duration are pricing parameters
Check whether an endpoint supports a given aspect ratio, resolution and clip length and whether each affects billing. Audio generation, sound synchronization, reference images and upscale passes can introduce distinct costs. Providers change models and render quality names frequently: save exact route/model IDs rather than a marketing family label such as ‘fast video.’ A published $/second for 720p should not be pasted onto 1080p work without evidence.
Operational spending controls
Before submitting a batch, compute a worst-case maximum based on allowed retries and route each job to a spend budget. Stop retry loops when the same prompt fails structurally. Persist job identifiers, initial settings, billed units and completed assets. Consider retention and egress where generated video is stored externally; a cheap render that triggers extensive manual editing is not necessarily cheap delivery.
Alternatives
Image-to-video shots built from approved visual assets, motion graphics, stock footage with appropriate rights and conventional editing can lower the amount of AI-generated footage required. Compare workflows on cost per accepted scene and continuity, not on headline clip pricing alone. Deep creative workflow advice belongs to AI 04; this page is a metering and budgeting guide.
A production ledger example
For each shot record scene ID, prompt version, approved reference-image ID, intended seconds, requested output resolution, actual billed seconds and acceptance decision. A clip that is technically generated but unusable for continuity still belongs in the cost denominator. When the vendor bills generated duration, not retained montage duration, trimming a bad frame in an editor does not refund the discarded seconds. For credit bundles, record the exact credit burn per mode separately; never assume an eight-second render costs exactly twice a four-second render without official evidence.
Financial gates before a large batch
Render a few representative scenes covering motion, text overlays and character continuity, then measure the observed acceptance ratio for that project’s standards. Project costs from the distribution, not from someone else’s advertised showcase. Budget a hard stop for retries on known-bad prompts and require review before upgrading resolution across all scenes. Where input assets are bought, licensed, rendered or upscaled externally, include those costs in the same scene ledger. A dollar-per-output-second headline is only one part of a finished documentary’s cost.
Decision checklist
- Identify the specific plan or API endpoint, model version, region and currency before pricing a job.
- Measure input, output, caching, generated media and failed-job units separately.
- Check the official pricing page again before procurement, especially if the SKU has no effective-date notice.
- Keep source evidence and a timestamped price ID; do not treat unknowns as zero.
Pricing evidence boundary
This cost discussion concerns Multi-vendor. Reproduce a quote only with its exact model or plan identifier, chargeable unit, billing region, processing mode and timestamp; a changing vendor rate card is not a permanent price guarantee. If any one of these dimensions is missing, keep the associated rate and worked scenario under editorial review instead of filling the gap with an estimate.
Sources & Further Reading
- Gemini Developer API shows tier/model/context/modality-specific rates and cache storage — official first-party source, accessed 2026-10-25; original effective date not specified in source.
- fal model-route output unit and GPU-hour list tariffs — official first-party source, accessed 2026-10-25; original effective date not specified in source.