One endpoint, 77 models
Every model in the catalog runs through the same request shape. Switching from one video model to another means changing a string, not rewriting an integration. The API documentation lists every field and response.
What it is for
Bulk product photography for a catalog. Video variants for paid social. Generation inside your own product, where users never see our interface.
Claude, ChatGPT and other agents
An MCP server and an OpenAPI schema ship in the package. Claude Desktop and Cursor get five tools. ChatGPT gets the same five as Actions. Setup is in the docs below.
Estimate before you spend
Send a model, a duration and a resolution to /api/estimate/. You get the exact credit cost back, plus a dollar figure and the settings the model accepts. It holds nothing and needs no balance, so a planner can price a whole batch first. If a model is paused or retired, the answer says so and names the replacements.
How billing works
API calls draw from the same credit balance as the interface. There is no separate API wallet and no separate rate card. What a run costs in the browser, it costs from code.
Who gets API access
Every account that bought a credit pack in the last 12 months, from Lite to Enterprise. Accounts with only the starting credits do not get keys. Programmatic traffic is many times heavier than manual use, so each key has its own rate limit. A data processing agreement and higher limits are part of a custom contract.
Authentication
Every request carries a key in the Authorization header. Keys are shown once, at creation, and stored only as a hash. A leaked database does not open your account.
Limits
Sixty requests per minute per key by default, higher under a custom contract. Exceeding it returns 429 with a Retry-After header rather than dropping the request silently.
Getting started
Buy any credit pack and create a key in your account, under API keys. Access stays open for 12 months after each purchase, like the credits. For larger volumes, contact us about a custom contract. Documentation is at the link below.
When an API is the wrong answer
If you generate a handful of images a week, the interface is faster than writing code. The API earns its keep at volume, in pipelines, or when generation has to happen without a person watching.
Failures and what they cost
A failed job releases its credits automatically, which matters more in a pipeline than in a browser. A loop that retries on error does not quietly drain a balance, because the failed attempts were never charged.
Rate limits and what happens at the edge
Requests are limited per key rather than per account, so a runaway script cannot starve your other work. When a model is paused for a cost problem, the endpoint returns a clear error with replacements. It does not fail silently.
Keys, and how to lose one safely
Issue a separate key per integration. Revoking one does not touch the others, and your account shows requests, runs and credits for each key. That is the difference between an incident and an afternoon of guessing.
Choosing a model in code
Fetch the models endpoint and pick by unit and rate rather than by name. Names change between versions. A hard-coded slug is the most common reason a working pipeline breaks three months later.
Testing without spending
Draft resolutions and the cheapest models exist for this. Run your integration end to end on a one-credit image model first. Switch the slug once the plumbing works. Debugging on a video model is an expensive way to find a typo.
