Read the catalog
workflow=api (the default) for API modes, or scriptToVideo for modes
available in the script-to-video workflow. Unsupported workflows return HTTP 400.
The response contains schemaVersion, endpoint, workflow, defaultModel,
and models. Version 1 describes visible models on the standard video task
endpoint. Refresh this response after a model update instead of keeping your own
model list. Account permissions, input validation and available balance are
checked separately when generating.
Build a request from a mode
- Select a model and one of its
modes. - Render its
controlsusingtype:select,boolean,numberortext. ApplydefaultValue,required, bounds andreadOnly. Numeric controls allow decimals whenintegerisfalse. - Apply matching
valuesWhenconstraints when another control changes. Each rule’swhenkeys must all match.valuesByResolutionsupplies the same behavior for resolution-dependent choices. Review changed settings again. - Collect the images, videos or audio described by
mediaSlots, respectingrequired,multiple,max, modelmediaLimitsand modematerialRules. - Place each control at its
requestPath. Without an explicit path, usecontrols[key]. Place media atassets[requestKey ?? key]; a slot withrequestMultiple: falseexpects a single URL. - Submit the canonical body to the returned
endpointwith your API key.
negativePrompt selection maps to
providerOptions.google.negativePrompt, while duration remains
controls.duration. Image references map to assets.referenceImages.
translations provide localized labels, help and enum display names.
Discovery covers form-compatible choices. Additional native parameters and
legacy request formats documented for each model remain available. Modes may
have different workflow availability; for example, Veo video extension is
API-only. Actual media bytes are validated before upstream submission, so
metadata discovery does not guarantee that a particular file will be accepted.