Prime Big Deal Days product videos: cap the spend per run

For a deal-day batch of product videos, set generation_spend_cap_usd on every item of a Format bulk run and keep concurrency small. The cap is that run's ceiling, up to the platform maximum of $500, and the effective cap comes back on every receipt, so a bad prompt cannot spend without limit.

Amazon's release, read 2026-09-30, says Prime Big Deal Days returns October 6-7 and that new drops launch three times daily, at midnight, 8 a.m. and 1 p.m. PDT. Those drop times are the moments you want finished creative ready, so queue the batch before the first one.

How does the per-run cap work?

Per the Calling a Format docs, omitting the field inherits the Format's cap, a number up to 500 is used as given, null runs at the $500 maximum, and 0 or anything above 500 is a 400. A cap lifts or lowers a ceiling; it does not guarantee a run fits under it, so size it from a trial run.

How do I read what a run actually spent?

Each receipt carries usage.generation_spend_cap_usd_micros and, next to it, usage.billable_amount_usd_micros, which is what the run spent against that cap. Read both after a trial item and set the batch cap just above the spend you saw.

How wide should the queue run?

concurrency takes an integer from 1 to 16; anything else is 400 invalid_request. A low number spreads spend over time, and you can stop the rest by cancelling children that have not started. Plan limits on generation concurrency are separate; see queue full vs concurrency full.

{
  "concurrency": 2,
  "items": [
    {
      "instruction": "Vertical 9:16 product clip for the Oct 6 drop",
      "input": { "product_url": "https://shop.example.com/p/101" },
      "generation_spend_cap_usd": 25
    }
  ]
}

What should I run before a large burst?

Sume's MCP docs say to prefer generation_admission_preview and/or dry_run before expensive bursts. Preview first, then send the queue with an Idempotency-Key so a retried request does not duplicate it. Prices follow the API pricing page.