Billing FAQ
The questions everyone asks, answered straight.
"I just ran something and my balance didn't change"
Two possibilities, both normal:
- It's still running. Charges post when work finishes, not when it starts. While it runs, the amber ⏳ In progress estimate on the Usage page shows what's accruing.
- It ran inside a workspace session that's still open. Sessions bill once, at close. Close the session and the charge appears within a minute.
If a run shows finished and nothing appears after a hard refresh and a minute — contact support via the in-app chat; every charge is individually auditable on our side.
"My video job takes 30+ minutes — will it get cut off?"
Two different clocks, know both:
- API/deployment runs obey your machine's Workflow timeout (Machine → Settings). If a run hits it, it's cancelled — and billed for the time it used. For long video jobs, raise the timeout before running: options up to 2 hours are available.
- Workspace sessions use the session timer instead — extend it live from the timer widget, no hard cap surprises.
If you're generating long/high-res video via the API, set the machine's Workflow timeout to comfortably exceed your longest expected job. A timed-out run bills for the time it consumed but delivers nothing.
"Why did a failed run cost money?"
GPU time was genuinely consumed — the model loaded and computed until the failure. You're billed for elapsed seconds only, never the projected full job. A run that dies in 10 seconds costs pennies or less.
"What happens at $0?"
Your running job always finishes — we never kill in-flight work over billing, even if it takes you below zero. After that, new runs/sessions are refused until you're positive again (top-up, next monthly grant). Negative balances are never auto-charged or invoiced.
"Do my credits expire?"
- Included monthly credits — reset each billing cycle (use them or lose them).
- Top-up credits — never expire. Spent after your included credits.
"Session vs run — which is cheaper?"
- Iterating rapidly (tweak → run → tweak)? A session is usually cheaper — the model stays loaded between runs, so you skip repeated cold starts. Close it when done: sessions bill wall-clock.
- Fire-and-forget single jobs? API runs — you pay exactly for the job, idle time is impossible.
"How do I estimate a job before running it?"
seconds × the GPU's per-second rate from the GPU pricing table. The pricing page has a simulator, and the first real run tells you the exact number — check its cost on the Usage page, and future runs of the same workflow will match it closely.
"Where do I get invoices / change my card?"
Plans page → Manage → Stripe billing portal. Invoices, payment methods, cancellations — all self-serve.