evaluation · consideration

Budget Studio credits for video-model experiments

Plan proof runs and stop conditions before high-cost video generation.

Define one decision the experiment must answer. Examples include whether a vertical five-second hook keeps the product readable, whether an approved still survives restrained motion, or whether a premium campaign beat is worth testing after the lower-cost concept is locked. Record the model path, aspect ratio, prompt version, input asset, slot limit, and review criteria before submitting.

Use pass/fail criteria that can be inspected: the required subject remains visible, camera motion follows one direction, important text is added later rather than judged inside generated pixels, the ending supplies an editable beat, and the output avoids a named defect from the prior run. Stop when the decision is answered, the reserved slot count is reached, or two consecutive attempts fail the same criterion without a new hypothesis. This prevents a vague request for improvement from turning into unbounded paid calls.

Start with 20 free credits

Model specimen

Budget from the current 42 / 42 / 96 catalog

The current reader-facing catalog prices Kling 2.6 Pro at 42 credits, Kling 2.6 Pro Motion at 42 credits, and Veo 3.1 Fast at 96 credits. Use those figures for page-level planning, then call `list_models` immediately before execution and use the current catalog record returned to the connected account.

These prices make one Kling text-to-video run and one Kling Motion image-to-video run equal-cost choices at 42 credits. A Veo run costs 54 credits more, or about 2.29 times either Kling path. That arithmetic is a budget comparison, not evidence of quality, speed, reliability, or suitability. A completed generation that misses the brief should still be treated as consumed experiment capacity unless a current account response says otherwise.

Kling 2.6 Pro: 42 creditsCurrent catalog price; confirm with list_models before execution.
Kling 2.6 Pro Motion: 42 creditsCurrent catalog price; confirm with list_models before execution.
Veo 3.1 Fast: 96 creditsCurrent catalog price; confirm with list_models before execution.

Sources

Prompt anatomy

Turn the credit envelope into experiment slots

After retrieving the current account catalog, calculate the maximum slots as available credits divided by the returned per-generation cost, rounded down. Write both inputs beside the result so another reviewer can reproduce the arithmetic. The published 42 / 42 / 96 figures are the current planning baseline, but the value returned to the account at execution time controls the actual spend; do not substitute direct provider dollars or a remembered price.

Allocate those slots before generating. A bounded plan can reserve roughly two-thirds for distinct proof ideas, one slot for a controlled revision of the strongest proof, and the remainder as contingency. Those proportions are editorial guidance, not Studio defaults or evidence that a fixed number of attempts will produce acceptable work.

If two models are under consideration, calculate each independently from the same opening balance and its own returned cost. A lower-cost path may support wider exploration, while a higher-cost path may leave little or no rerun capacity. That is a budget observation, not a quality ranking. Do not recommend either model without matched outputs and a documented acceptance test.

Sources

Credit gauge

Do not confuse Studio credits with direct provider billing

The underlying fal pages use a different billing unit from Studio credits. Kling 2.6 Pro documents per-second provider pricing and five- or ten-second requests, while Veo 3.1 Fast documents per-second pricing and four-, six-, or eight-second requests. Those figures describe direct provider endpoints. They do not establish the current Studio credit charge.

The public @offlinecreator/mcp 0.1.2 package routes `list_models` to the Studio models API and describes it as returning launch models with credit costs and workflows. Use the complete runtime model record, including any model-specific options it returns, as the execution input. Do not multiply a fal per-second price by an assumed duration or infer that every provider option is selectable through Studio.

Sources

Related circuit

Open the model directory to refresh the launch catalog before spending. Use the source-image requirements guide when the experiment starts from an approved still, or the Veo guide only after a premium text-to-video slot has a fixed brief. These routes move from budget to the next operational constraint without turning this page into a general model ranking.

Next action

Run and document a bounded Studio experiment

First call `list_models` immediately before the experiment and record the full returned model entry: model ID, workflow, credit cost, and any model options present. Call `get_credits` to capture the opening balance. The published 0.1.2 package exposes both tools, and its server implementation describes them as the launch catalog with credit costs and workflows and the current account balance.

Second, submit one proof using the chosen model and a versioned prompt. Review it against the written criteria before changing anything. If it fails, name one defect and change one variable: prompt wording, ratio, source image, or motion instruction. Do not change model, framing, subject, and camera move together because the result will not explain which change mattered.

Third, log opening balance, model ID, returned cost, selected options, generation ID, prompt version, input version, observed result, decision, and closing balance. Compare the balance delta with the returned cost. If they disagree, stop the experiment and preserve the response as a billing-verification issue rather than estimating the next call.

Cancel only when the job is still reserved and has not been submitted to the provider. The published 0.1.2 `cancel_generation` contract limits that action to a reserved generation before provider submission and describes a credit refund. Do not plan on cancellation as a way to recover credits from a completed but creatively weak output.

Sources

Start with 20 free credits

Canonical plate

Evidence boundary and refresh rule

The current planning prices are 42 credits for Kling 2.6 Pro, 42 for Kling 2.6 Pro Motion, and 96 for Veo 3.1 Fast. A preserved August 9 snapshot in the claim ledger records that the public pages previously displayed 60, 60, and 600; that dated observation is historical only and must not be presented as a current price. Always advise `list_models` because the connected catalog remains the execution-time authority.

The sample slot method and stop conditions are editorial guidance. They are not evidence of output quality, success rate, typical retry count, refunds for completed but unsuitable outputs, or customer outcomes. Recent community material did not meet the page's evidence standard, X was unavailable, and Reddit coverage was partial.

Sources

Budget from the current 42 / 42 / 96 catalog

Turn the credit envelope into experiment slots

Do not confuse Studio credits with direct provider billing

Continue with the constraint that controls the next run

Evidence boundary and refresh rule