Plan AI video duration and credit cost

Choose the duration path before writing the concept

Treat duration and credit cost as constraints at the start of an AI video brief, not as details to discover after the prompt is written. For a bounded comparison, this draft uses a five-second Kling 2.6 Pro scenario at the current 42-credit Studio charge and an eight-second Veo 3.1 Fast scenario at the current 96-credit charge. Five seconds and eight seconds are documented options for the disclosed provider endpoints; the public Studio catalog does not publish duration, and the public MCP `generate` schema does not let a caller select it. The scenarios are planning references, not verified Studio duration controls.

Begin by writing the shot's purpose in one sentence. If the purpose can be shown through one subject action and one camera move, test whether it fits the five-second editorial scenario. Consider the eight-second scenario only when the concept genuinely needs more time for a readable setup, movement, and settled end frame, and when the higher catalog charge has been approved. If exact duration is a hard requirement, do not authorize generation from this draft alone: the authenticated model response was not queried. If the idea requires several locations, cuts, claims, or sequential actions, split it into a storyboard or editing plan instead of packing more instructions into one generation.

Compare the current model and cost envelope

Use `list_models` immediately before generation to confirm the available model IDs, workflows, and published credit costs, then call `get_credits` to read the account balance. The public MCP implementation describes those two tools as read-only discovery steps. It does not establish that `list_models` returns duration. Its `generate` schema accepts `modelId`, `prompt`, optional `aspectRatio`, and optional `wait`; it does not publish a duration argument. Never add an unsupported `duration` field to an MCP call.

The underlying fal documentation is broader than the public Studio surfaces reviewed here. The disclosed Kling endpoint documents five- and ten-second options, while the Veo endpoint documents four-, six-, and eight-second options. Those provider choices establish provider duration envelopes, but they do not prove which duration Studio uses or that any duration is selectable through OfflineCreator. The current Studio page publishes account credit charges, and the public MCP schema publishes callable fields. Keep those evidence types separate; this research did not query the authenticated model endpoint or verify a live Studio duration value.

Kling 2.6 Pro

Editorial scenario: 5 seconds; current Studio charge: 42 creditsFive seconds is documented by the disclosed provider endpoint. It is not presented here as a caller-selectable Studio MCP control.

Veo 3.1 Fast

Editorial scenario: 8 seconds; current Studio charge: 96 creditsEight seconds is documented by the disclosed provider endpoint. The public Studio MCP schema does not expose that provider duration field.

Budget proof runs before variations

Separate the cost of proving a shot from the cost of producing a set. One Kling proof costs 42 credits at the current published rate; two controlled Kling attempts cost 84 credits, and three cost 126. One Veo proof costs 96 credits; two cost 192, and three cost 288. These are arithmetic planning totals based on the cited per-generation catalog charges, not quoted currency prices, subscription estimates, refund promises, or predictions about how many attempts will succeed.

A useful approval note names the model, provider-documented duration scenario, number of authorized attempts, maximum credits, and stop condition. Example: `Kling 2.6 Pro / editorial 5-second scenario, not an MCP control / up to 2 attempts / 84-credit ceiling / stop after one accepted shot or after the second rejection.` For the higher-cost comparison: `Veo 3.1 Fast / editorial 8-second scenario, not an MCP control / 1 attempt / 96-credit ceiling / no automatic retry.` Check the live balance before each paid action instead of assuming that a balance captured earlier still applies.

Do not budget every possible variation at once. Fund one proof against explicit acceptance criteria, inspect the complete output, and authorize the next attempt only when a single controlled change can answer a useful question. A second run that changes subject, camera, lighting, pacing, and ratio together spends credits without isolating why the first result failed.

Fit the beat sheet to five or eight seconds

For a five-second shot, reserve roughly one second to establish the subject, about three seconds for the single action or camera move, and the final second for a stable exit frame. This is an editorial timing template, not a model guarantee. It works best for one readable change: a product light turns on during a slow push-in, fabric moves once in a controlled breeze, or a subject turns toward camera and settles.

For an eight-second shot, do not automatically add more events. Use the extra time to improve readability: establish for two seconds, carry one action through the middle four seconds, then hold the result for two seconds. The longer path is justified only when those additional seconds serve the final edit. It is not justified merely because the concept sounds more important, and the higher credit charge does not establish stronger coherence or better prompt adherence.

Write the timing next to observable review points: what must be visible at the opening, when the action should begin, where the camera should stop, and what the final frame must preserve. If the required action cannot be described without phrases such as “then,” “after that,” and “cut to,” split the idea into separate shots and budget each one rather than pretending it is a single continuous clip.

Run a model, balance, and approval preflight

Use a fixed preflight order. First, call `list_models` and select the exact current text-to-video model ID. Second, record only the fields established by the public evidence: workflow and credit cost. Third, call `get_credits` and compare the balance with the one-attempt cost and the approved ceiling. Fourth, show the operator a compact plan containing model, provider-documented five- or eight-second scenario, ratio, prompt revision, attempt count, maximum credits, and an explicit note that duration was not verified as a Studio control. Generate only if that uncertainty is acceptable.

The public MCP code supports the cost boundary: `list_models` returns launch models with costs and workflows, `get_credits` returns the current balance, and `generate` starts a generation. It does not support comparing or validating duration through `list_models`. If the live workflow or credit cost differs from this page, stop and revise the budget. If exact clip length matters, obtain separate current first-party Studio evidence before spending; do not infer it from provider documentation or invent a caller-supplied override.

After submission, keep the generation ID with the approved budget line. A completed output still needs human review; a failed, timed-out, or rejected creative result is not permission for an unbounded retry. Return to the approval ceiling and decide whether the remaining credits cover a meaningful controlled change.

Choose the next guide from the planning decision

Use the aspect-ratio guide when placement determines whether the shot should be landscape, vertical, or square. Move to the generation-manifest guide when the editorial duration scenario and credit ceiling are documented and the team needs a durable record of model, prompt, ratio, source input, job ID, and selection. Return to the creative-workflow directory when the concept is not yet narrow enough for one paid shot.

Keep the plan inside the evidence boundary

This page owns one decision: choosing a provider-documented duration scenario and setting a credit ceiling before a video generation. It does not verify Studio's generated clip length, compare visual quality, benchmark model reliability, promise that a shot will pass review, convert credits into a universal currency price, or claim that provider duration options are available through Studio. Current product facts come from the public Studio catalog and public MCP implementation; endpoint duration options come from fal's current documentation.

The page remains an unreviewed research draft. The authenticated model endpoint was not queried because production credentials are outside the research boundary, and no paid generation was run. Recheck `list_models` only for current model ID, workflow, and credit cost; recheck the balance separately. Do not treat either response as duration evidence unless its actual current payload explicitly supplies it and is preserved for review. Under this page's monthly freshness cadence, the dated product claims should be reverified by September 8, 2026.