Generate media from JavaScript

Generate images, videos, voices, music and sound effects with any catalog model from Block code, paid with StillMade credits after one approval.

A JavaScript Block can generate images and videos with any model in the shared catalog (the same models the app's own Canvas and Image Panels use), speech in any catalog voice, music and sound effects by returning a declared media.generate action. Declare it in the manifest:

"permissions": {"project": [], "actions": ["media.generate"], "network": [], "filesystem": [], "secrets": []},
"outputs": {"request": {"type": "json"}, "media": {"type": "asset[]"}}

Return the action from your code. Each item is an image, video, speech, music or sfx item with a prompt and optional settings. Image and video items may also set references (images), firstFrame/lastFrame (video), negativePrompt and seed. Audio items take only settings:

• speech: the prompt is the text to speak (up to 3,800 characters);

settings: {voice, speed} with a catalog voice id and a speed from 0.25 to 4.

• music: settings: {durationSec}, 5 to 240 whole seconds (default 30).
• sfx: settings: {durationSec}, 0.5 to 22 seconds in tenths (default 4).

return {request: {
  schemaVersion: 1, kind: "stillmade.media-action", operation: "media.generate",
  resultPort: "media",
  request: {items: [
    {key: "hero", operation: "image", prompt: input.prompt,
     settings: {model: "nano-banana-pro", aspectRatio: "16:9", resolution: "2K"},
     references: input.styleImages},
    {key: "clip", operation: "video", prompt: "Slow push in on the harbor",
     settings: {model: "kling-v3-turbo", duration: 5, aspectRatio: "16:9"}},
    {key: "intro", operation: "speech", prompt: "Welcome back to the show.",
     settings: {voice: "adam", speed: 1}},
    {key: "bed", operation: "music", prompt: "Warm lo-fi piano", settings: {durationSec: 30}},
    {key: "whoosh", operation: "sfx", prompt: "A fast whoosh", settings: {durationSec: 1.5}}
  ]}
}};

StillMade turns every item into one exact generation request and asks the server for one maximum price for the whole batch, using StillMade's own prices. The user sees that maximum and approves it once; credits are reserved before any provider call, each call is settled from provider evidence, and unused credits are returned when the batch closes. The approval is bound to this Block's exact content digest, so changed code needs a new approval. A Block cannot name a price, a provider, a key or another Block. References and frames must be images saved in the user's own library. Up to 50 items per action.

When it finishes, the declared resultPort receives an array of {key, kind, assetId, versionId, url} in item order (kind is image, video or audio); failed items are listed separately and are not charged beyond what the provider actually received.