Veo 3.1 audio block: not charged. What happens on a failed Sume job?

If Veo 3.1 blocks your video because of an audio problem, Google says you are not charged. Its Veo guide states that Veo 3.1 will sometimes block a video from generating because of safety filters or other processing issues with the audio, and that you will not be charged if the video is blocked. Sume works from a different rule: it reserves the estimated cost when it accepts a job and releases or refunds that reservation if the job fails.

Google's line is from its Gemini API: Generate videos with Veo, read 2026-10-02. Sume's side is from Generation admission and Jobs and results. Sume's current model catalog does not list a Veo id, so the Sume half applies to the models it does list, such as Gemini Omni Flash 1.1.

What exactly does Google promise?

Only the blocked-for-audio case. The sentence covers Veo 3.1 videos blocked by safety filters or other audio processing issues. It does not describe other failures, a timeout, or a result you dislike. Google's page does not say what the response looks like, so handle any non-success result in your code as no video.

How does Sume handle a failed job?

Sume's admission page says paid generation uses public USD estimates. At submit time it reserves the estimated amount when the request is accepted; successful completion captures the reserved usage; failed jobs and failed queue admission release or refund the reservation where applicable. The core workflow page puts it as capture on success and refund on failure or cancellation before capture.

Cancelling works only before generation starts. Once it has started, cancel returns 409 job_generation_already_started and the job completes or fails normally.

How do I confirm it on my account?

Read the job's terminal status, then read the usage ledger. Sume's API reference describes GET /v1/usage as the ledger of entries such as reservations, captures, refunds and top-ups, so a failed job should show a reservation followed by a release or refund. Do the same check after your first failed test job before you trust a budget model.

What should I do in a batch?

Three habits keep a batch predictable.

• Poll to a terminal state; do not resubmit a failed job without a fresh decision.
• Reuse the same Idempotency-Key only for a retry of the same request.
• Track the ledger, not just job counts.