Pricing and billing

Every price is a flat USD number you can total before a task starts. An image is priced per output, a video per second, and the analysis tools per call. Creating a task holds the estimate on your balance. Completing it charges only what succeeded and releases the rest.

There is no subscription and no minimum spend. You prepay credits in the wallet. The minimum top-up is $5, with no Mynth fee on top. An idle account is never billed.

Where prices come from

Providers bill by megapixel, by token, or by GPU time. Mynth publishes one number per model, usually the provider's own price. Where it is higher, the difference covers routing, retries, and delivery, and it is already included.

The catalog carries prices as decimal strings in USD:

The catalog is at mynth.io/models, the pricing table, GET https://api.mynth.io/models, /models.json, and npx @mynthio/cli models list. None of them need an API key.

Image price

per image  = per_image[scale] + per_input × number of inputs
task       = per image × successful images

scale is 4k for a _4k size and base otherwise. Nothing else changes the price: not prompt length, pixel count within a scale, output format, webhooks, destinations, which provider ran the job, or how many retries it took.

For example, with per_image.base at $0.03 and per_input at $0.03, a request with count: 2 and two reference images holds $0.18 and charges $0.09 per image that succeeds.

A _4k size on a model without 4k is rejected at create with 400 validation_error, so nothing is held. With model: "auto" the model is not known yet, so the task fails with capability_not_supported instead, and a failed task costs nothing.

With model: "auto", the hold is a flat $0.20 per image because the model is not chosen yet. On completion Mynth charges the published price of the model in result.model and releases the rest. See the model field.

Video price

video = (per_second[tier] + audio.per_second) × duration

tier is the resolution in size. The aspect ratio does not change the price. An omitted tier, duration, or audio uses the model's default, and the estimate includes it. audio.per_second applies only when the catalog lists it. The video models available today include audio in per_second.

Tool prices

Magic Prompt is included in the generation estimate, so if it gets a price later, the estimate will show it.

Before you send

POST /image/generate/estimate and POST /video/generate/estimate take the same body as generate. They validate it and return the price without creating a task or holding anything. Estimate cost has the request and response.

Balance

available = balance − reserved

balance     credits on the account
reserved    held by tasks that are still pending
available   what the next task can hold

GET /balance returns all three as decimal strings. It needs a key with the manage scope. npx @mynthio/cli balance prints them.

{
  "data": { "balance": "12.50", "reserved": "0.18", "available": "12.32", "currency": "usd" }
}

Holds and charges

create task ── hold estimated_cost against available
  │
  ├─ completed ── charge each successful output, release the rest of the hold
  └─ failed ───── release the whole hold, charge $0

• The create response returns the hold as estimated_cost.
• If available is below the hold, the create fails with 402 insufficient_balance and nothing is queued.
• If the key's spending limit would be passed, it fails with 402 spending_limit_exceeded.
• Only successful outputs are charged. If three of four images fail, you pay for one.
• The amount charged is cost on the finished task. It is null while the task is pending and when it failed.

A task still pending 24 hours after its last update is failed with task_expired by a sweep that runs daily at 02:00 UTC, and its hold is released. The sweep looks at tasks last updated within the past 7 days. If reserved stays high after that, email mynth@mynth.io.

Refunds

Credits are for spending on Mynth and are not paid back as cash. The terms cover refunds when a charge was our fault. The wallet lists each top-up and each charge with what it paid for.

Next steps

• Estimate cost: price a request before you create it.
• Tasks: where cost lands, and partial success.
• Choosing an image model: comparing prices across models.