Credits are the least interesting part of the product and the part people ask about most, which usually means the explanation is not good enough. Here is the whole model.
Compute is the unit, not the request
A request is not a fixed amount of work. A four-second clip at 720p and a twelve-second clip at 1080p are the same API call and differ by roughly an order of magnitude in what they cost to produce. Pricing them identically would mean one of the two is subsidising the other — and in practice it means the cheap request funds the expensive one, which is a tax on everyone doing small work.
So the unit is compute, expressed as:
| Output | Priced by |
|---|---|
| Image | Resolution and step count |
| Video | Duration, resolution and frame rate |
| Audio | Duration |
| Edits | The size of the region processed |
Why the estimate comes first
Every generation is quoted before it runs. Not as a courtesy — as a design constraint. A pipeline that can charge you an amount you did not see is a pipeline you cannot safely automate, and automation is most of what this product is for.
The consequence worth knowing: for a per-second app, the quote is zero until the source duration is known. That is not a free generation. It means the input has not been measured yet, and the number you should be reading is the per-second rate, not the total.
Concurrency is separate from credits
The most common billing confusion is between two limits that feel the same and are not:
- Credits are how much work you may do in total.
- Concurrency is how much of it may happen at once.
Running out of credits stops you. Running out of concurrency queues you. A plan upgrade that doubles your credits and doubles your concurrency is doing two unrelated things — and if your problem is that a batch of forty renders takes all afternoon, the credits half of that upgrade is not what fixes it.
Unused credits
They roll over for the life of the subscription. There is no monthly expiry, because a monthly expiry turns a production tool into a thing you have to remember to use, and nobody has ever done better work under that pressure.