Help
How you are charged for generations, what counts as successful, and where to check your spending.
Image, video and audio generation charges settle when a run completes successfully. Confirmed failed or cancelled generation attempts are not charged. Workflow estimates reflect typical usage; the final charge follows actual usage. A timed-out or unconfirmed run can still finish and be charged.
While a generation is in progress, its estimated cost is put on hold. Some runs include a small temporary safety buffer above the estimate. Your balance doesn't drop yet, but the hold counts against what you can start next, so you may see Insufficient balance while other runs are in progress or when your balance covers the estimate but not the full hold. The buffer is not an extra final charge. On success you are charged; on confirmed failure or cancellation the hold is released. If the billing outcome remains unconfirmed, the hold can remain until it is reconciled.
Describe and Transcript are paid processing tasks. Completed processing can be charged even when its text is empty or unusable. A failed status alone does not prove that nothing was charged; check the billing state and Generation History.
The Assistant needs an active subscription. Each successful Assistant reply is charged from your balance. Paid generations it proposes use your generation balance and wait for your approval.
With Auto-approve paid runs on, the Assistant starts paid generations without asking until the per-session limit is used. Pick $5, $10 (the default), $25, $50 or Off. Off removes the session limit while auto-approval stays on; server limits still apply to each run. To restore approval prompts, turn off the Auto-approve paid runs switch. The session limit counts only the estimated price of generations it approved, not the Assistant's own replies.
Stopping the Assistant does not cancel a generation that has already started. It keeps running and is charged if it finishes successfully.
Most balance updates happen within seconds. In rare cases where the generation takes unusually long or the status update gets delayed, a background process reconciles stuck tasks, typically within an hour. Some generations can take longer on the provider side, and the hold stays in place until the final result comes back.
Your plan sets how many generations can run at the same time: no active plan 2, Starter 10, Creator 20, Pro 50, Enterprise the highest.
Board runs in this tab wait in separate queues for different types of generation. All runs in this tab, other tabs and connected tools share your account's active-run limit. Another run can fill the available slots, so a request can still be refused with "Concurrent generation limit reached".
Supported Shift-batches show a price review before starting. Workflow batch prices are estimates of actual usage, not spending caps; the final charge can differ. Sequential Lip Sync runs one character at a time and does not support Shift-batching.
A generation is "successful" when it completes and delivers output, whether that's an image, video, or audio file. This is based on whether output was delivered, not on visual quality.
If you receive output that doesn't meet your expectations (artifacts, low quality, unwanted results), that is still billed because the processing was completed and output was delivered.
Runs confirmed failed or cancelled without output are not charged. A timeout or a missing result alone does not confirm either outcome.
The cost shown per board only counts what is currently on that board. Deleting items, clearing a board, or working across multiple devices will make the board total lower than your actual spending.
Board totals can also differ if items were duplicated, moved between boards, or mixed with locally edited versions. A board is a snapshot of what is visible now, not a full spending history.
For accurate numbers, use Generation History for per-generation charges and Account for your current balance.
I deleted a generated video from my board. Will I get a refund?
No. The charge is applied when the output is delivered. Deleting the result from your board afterward does not reverse it.
A generation failed but my balance looks lower. What happened?
The failed task's hold is released, but other generations still in progress keep their own holds until they finish. If it still doesn't add up after everything settles, check Generation History first. If you think a failed generation was billed incorrectly, contact support with the details.
The result quality was bad. Can I get credits back?
If output was delivered, the generation is not refundable based on quality alone. AI results vary and different prompts, models, or settings can produce better outcomes. Refund requests are still reviewed for billing errors or unresolved technical issues.
My board shows less spending than what I was actually charged.
Board costs only count items still on that board. If you deleted outputs, switched devices, or cleared browser data, the total won't match. Use Generation History for the full charge breakdown and Account for your current balance.