Credits you can actually predict

Image and video generation, priced before you press go.

Every generation is quoted by the server before it runs. You see the exact credit cost, the credits are held while the job works, and anything the job does not use comes straight back.

Quoted, not estimated

The price is computed server-side from a versioned pricing rule and held in a signed record. The number you approve is the number you pay.

Nothing is silently consumed

Credits are reserved, not spent. A failed or cancelled job releases them in full, and a partial result is charged pro rata.

Private by default

Outputs live in private storage and are delivered through short-lived signed links. There is no public gallery and nothing is shared.

A ledger you can read

Every grant, hold, charge, release, and expiry is an immutable line item. Your balance is reconstructible from it, entry by entry.

Providers are swappable

Models are product names, not vendor names. When something behind one changes, your history and your workflow do not.

Age-safe by design

A content policy runs before anything reaches a provider, so rejected requests never enter a queue or cost a credit.

How a generation works
  1. Choose a model and describe what you want.
  2. The server returns an exact credit quote, valid for a few minutes.
  3. Approving it reserves the credits and queues the job.
  4. Progress streams while a provider runs it.
  5. On success the reservation settles; on failure it is released.