POST /v1/{architecture}/generate, and its own page here listing every field it accepts. The same list, with each field’s tokens and defaults as JSON Schema, comes back from GET /v1/architectures, so a client can discover it at runtime.
Variants of a model (Plum and Plum Max, the Cherry line) are the model_id field of one architecture. Gems are the price of the default settings; other resolutions and durations cost more or less, and the charge for a request is in its response.

