Why Did One AI Video Generation Show Both a Charge and a Refund?

A charge plus a partial refund on one AI video generation is normal settlement: a maximum-cost hold, then unused credits returned. Here's how to reconcile it.

GGoFaceless Team7 min read
Illustration of a charge leaving a wallet and a partial refund returning, representing AI video generation billing settlement.

A charge and a partial refund on the same AI video generation is normal billing settlement, not a billing error: the platform placed a maximum-cost credit hold before running a variable-duration job, then returned the unused credits once the job finished. Your true final cost is the charge minus the refund, and the pair of entries is how the ledger shows that math.

Key takeaways:

  • Runway documents a maximum-cost hold for certain variable-duration generations and edits, with unused credits returned after the job completes.
  • Final cost of a settled job equals the charge minus the refund; treat the hold as temporary balance usage, not spending.
  • A failed generation is handled differently: it receives a full return and the charge entry is removed entirely.
  • Runway's documentation lists Seedance 2.5 at 68 credits per generated second for 1080p and 20 credits per generated second for 480p, so resolution alone changes the hold size by more than 3x.

What is a maximum-cost credit hold?

A maximum-cost credit hold is a temporary deduction equal to the most a variable-duration generation or edit could possibly cost, taken from your credit balance before the job starts. Runway documents this pattern in a help article literally titled "Why do I see a charge and a refund for the same generation", which explains that certain variable-duration generations and edits are billed at their maximum possible cost up front, with unused credits returned after completion. The hold guarantees the platform that you can pay for the worst case; the refund corrects the ledger to the actual case.

The practical consequence: your available balance drops by the maximum, not by the real cost, for the duration of the job. If you run several jobs at once, each one holds its own maximum, which is why a balance that looks sufficient for your planned spending can still block the next launch.

How do I calculate the final cost after a partial refund?

Final cost equals the charge minus the refund, full stop. Runway's documentation states this directly: for a successful job with an unused-credit return, the final cost is the charge minus the refund. If a generation shows a 680-credit charge and a 340-credit refund, you spent 340 credits. The 680 was never spending; it was the hold.

A simple reconciliation habit for credit-based services:

  • Record the pair, not the parts. Log each completed job as one line: charge, refund, net.
  • Reconcile on net cost. Budgets and per-video cost tracking should use the net figure only.
  • Flag missing refunds. A charge with no matching refund after completion is the actual anomaly worth a support ticket.
Illustration of a credit hold leaving a balance and a smaller partial refund returning, settling level on a scale.
Illustration of a credit hold leaving a balance and a smaller partial refund returning, settling level on a scale.

Does a partial refund mean the generation failed?

No. A partial refund paired with a completed output means the generation succeeded and simply used less than its maximum. Runway's documentation draws an explicit line between the two cases: failed generations receive a full return and have the charge entry removed, while successful jobs keep the charge and refund pair as the settlement record. So the diagnostic is simple: charge removed entirely means failure; charge plus partial refund means success at a lower-than-maximum cost.

This is a different question from refund eligibility for a clip that generated successfully but is unusable for creative reasons. A defective-but-completed clip is a quality dispute, not settlement, and it follows a separate policy. If that is your situation, read up on getting AI video credits back for an unusable clip rather than interpreting your settlement entries as a quality refund.

Why might a job need more available credits than it ultimately costs?

A job needs more available credits than it ultimately costs because the platform cannot know the billable duration or work of a variable-length generation until the job finishes, so it secures the maximum first. Runway applies this to certain variable-duration generations and edits, where the final metered amount depends on what the job actually produces. The hold is collateral, not a price quote.

For creators, the operational effect is queue capacity. Two jobs that will each net 300 credits might each hold 680, so a 1,000-credit balance runs one job at a time, not two. If your batch sits idle, the hold math is often the reason; the same logic explains why an AI video batch queues instead of generating all at once.

How does output resolution change the credit rate?

Output resolution changes the per-second credit rate directly, and the gap is large. Runway's documentation lists Seedance 2.5 at 68 credits per generated second for 1080p and 20 credits per generated second for 480p. These are platform-credit rates, not dollar prices, and the page supplies no effective date, so verify current rates before budgeting. At those listed rates, a 10-second 1080p clip meters 680 credits against 200 at 480p, a 3.4x difference for identical duration.

Seedance 2.5 listed credit rate per generated second
1080p68credits/sec480p20credits/sec

The listed 1080p rate is 3.4 times the 480p rate for the same generated second.

Seedance 2.5 listed credit rate per generated second
1080p68credits/sec
480p20credits/sec
Source: Runway Help Center

The workflow implication: draft at the lower resolution when reviewing motion and composition, and reserve the high-rate renders for finals. That habit also shrinks every hold, freeing balance for parallel jobs.

Illustration of a checklist beside queued video frames and a credit gauge, representing pre-launch checks for batch generation.
Illustration of a checklist beside queued video frames and a credit gauge, representing pre-launch checks for batch generation.

What should I check before launching several variable-duration jobs?

Before launching several variable-duration jobs, check that your balance covers the sum of maximum holds, not the sum of expected costs. Runway's settlement model means each job temporarily deducts its maximum possible cost, so the pre-launch checklist is:

  • Balance vs. holds: total the maximum cost of every queued job and compare against available credits.
  • Resolution per job: confirm each job's output resolution, since the listed Seedance 2.5 rates differ by 3.4x between 480p and 1080p.
  • Refund timing: expect unused credits back after each completion, and do not count them as available until they land.
  • Records: keep the charge, refund, and net for every job. A simple production log, like the ones described in production records to save for every AI-assisted video, turns a confusing ledger into an auditable one.

One organizational note: an AI faceless-video platform such as GoFaceless (our own product) can be one way to keep drafts and final renders for a series in one place, since it produces script, voiceover, visuals, and captions in a single pass. But billing models differ across services, so verify whether any platform you use employs credit holds at all rather than assuming the charge-and-refund pattern is universal.

The bottom line

A charge plus a partial refund on one generation is the ledger showing its work: maximum hold in, unused credits back, net cost in the middle. Reconcile on the net, size your balance to the holds, and reserve actual refund disputes for failed or defective outputs.

If you want a simpler pipeline for the creative side, GoFaceless turns a topic into a finished short-form video with script, voiceover, visuals, and captions in one pass, with plans from $29/mo on the pricing page. Whatever tool you use, check its billing documentation before your next batch.

Sources & further reading

Keep reading

Ready to create your first video?

See examples made with GoFaceless. Create your own through a card-required one-video trial; after the trial, your selected plan begins unless you cancel.