generate_image, generate_video and generate_audio run a model on a prompt and, where the type accepts them, on library items. Each returns a job, and get_jobs returns the item when it is ready. A generation the model fails to produce is refunded.
Models
Every generation runs on a model. When none is given, the account’s default for that medium applies;manage_settings sets it. list_models is the catalogue: each image and video model’s exact settings, limits, accepted references and supported types, and which model is the default. type: "voice" lists the voiceover voices, and every voice speaks every language generate_audio accepts. Settings are validated against the catalogue before anything is charged. The catalogue changes as models are added and retired.
Types
Each model’s catalogue entry lists the types it accepts.
source_refs are the media the model works from. reference_refs are images for style or character consistency, or videos for motion, on models that accept them.
Price
Resolution, duration, aspect ratio, audio and count set the price, each taken from the call or from the model’s default.estimate_cost quotes a planned call from the same rates. Every charged call takes an idempotency_key; see Costs.
On a clip
Withtimeline_id and clip_id, a generation or an upscale works on that clip’s media, sends and prices only the stretch the clip plays, and the result takes the clip’s place when it is ready, with its position and look kept.
The result
A finished generation carries its prompt and settings on the item (get_items), which is the starting point for a variation. Readable text rarely survives video generation, so words are usually placed on a still first, or added on the timeline as text.
Related tools
upscale_media (image to 4K, video to 4K up to five minutes at a time), remove_background and reframe_image are charged in credits the same way and return jobs. extract_media, which takes a still or a stretch out of a file, is free.