POST /v1/videos; the body decides the path. Video models
declare modes such as text-to-video, image-to-video (a first_frame, sometimes a last_frame),
reference-to-video (reference_images, reference_videos, reference_audios) and video edit.
GET /v1/models/{id} lists each model’s modes with the required and optional inputs, allowed values,
resolution tiers and a ready-to-send example_request per mode. The pages under Video models
in the sidebar show the same, one per model.
Price depends on resolution and duration, so estimate with the same
duration_seconds and tier
you will send. Reference inputs take asset ids from Uploads or an earlier output; the
catalog’s duration and frame-rate rules are checked before any credit is reserved.
Model-specific prompting guides: Seedance 2,
Kling, LTX 2.