Submit with POST /v1/generate carrying a mode (text | image | video | audio), a prompt, and an optional model id from the catalog. You get a job id back; poll GET /v1/generations/{id} every few seconds until status is done (the result holds a media URL or text) or failed.
Inputs
init_image- a source image URL. Used for image edits, image-to-video, AND image-to-text on a vision model.last_frame- a target last frame for first-last-frame video.audio- a voice track URL (talking-head engines).target_video- footage to swap a face onto (face-swap engines).driving_video- a performance to copy (reenactment / character-swap).
See image-to-text (vision) and the video engines for worked examples.