Neural voiceover
Add a voice element with text and a voice. The renderer synthesizes the narration and can size the scene to fit it exactly.
elements[].type: "voice"
Voice, subtitles and images — declared, not produced
Neural voiceovers with word-timed subtitles and prompt-generated images — declared inline in the same JSON document.
Word-timed subtitles
Auto subtitles come from the speech synthesis itself, so each word highlights at the exact moment it is spoken — no transcription step.
Generated images
Give an image element a prompt instead of a URL. Seeded generations are reproducible, and repeated renders reuse the cache for free.
AI-drafted documents
Describe the video you want in a sentence and get back a valid document to refine in the editor — validated before you ever see it.
Deterministic renders
Same document, same output. Generation happens in a prepare step, so the render itself is a pure function of the document.
Metered fairly
Image generations are metered per generated image; cache hits and re-renders cost nothing extra.
{
"id": "voiceover",
"type": "voice",
"text": "{{script}}",
"voiceId": "Joanna",
"language": "en-US"
},
{ "id": "captions", "type": "subtitles" },
{
"id": "backdrop",
"type": "image",
"generate": { "prompt": "sunrise over a city skyline, cinematic", "seed": 7 }
}