Natiq-studio/voiceover

Use the connected Natiq MCP tools and follow the server's instructions for catalogue discovery, billing, asynchronous jobs and download URLs. If the server is unavailable, explain the connection problem; never pretend audio was generated.

Keep the narration conversational Algerian Darija, with short spoken sentences. Preserve product names and intentional French words. For high-stakes copy or uncertain regional phrasing, ask for a native-speaker review. Do not claim perfect Darija. Keep each scene independently editable, and keep one voice and delivery style unless a character change is intentional.

Before generation, create a scene manifest in the user's video project containing scene ID, narration, voice, purpose, country, speed, job ID, local audio path and measured duration. Save the job ID as soon as a submit returns. Compare narration AND voice/performance settings before reusing audio. If only visuals change, keep the audio. If the requested runtime is too short after generation, explain the tradeoff and shorten the script or extend the scene; do not silently cut speech.

Download successful clips into public/audio/ for Remotion (or the project's existing asset directory). Use stable names such as scene-01.wav. Signed URLs are temporary: store local paths and job IDs in the manifest, not signed URLs. Do not forward the API key to the audio host. Example, substituting the returned URL:

mkdir -p public/audio
curl --fail --location --output public/audio/scene-01.wav 'RETURNED_AUDIO_URL'

Use the existing project's frame rate. Round each scene up to whole frames and add any deliberate pause separately. For example, with scenes from the manifest:

import {Html5Audio, Series, staticFile, useVideoConfig} from 'remotion';

export const Narration = ({scenes}) => {
  const {fps} = useVideoConfig();
  return <Series>{scenes.map(scene => (
    <Series.Sequence key={scene.id}
      durationInFrames={Math.ceil(scene.duration_seconds * fps) + scene.pauseFrames}>
      <Html5Audio src={staticFile(scene.audio)} />
      {/* Put the scene's visuals here, inside the same Sequence. */}
    </Series.Sequence>
  ))}</Series>;
};

Set the composition duration to the sum of those exact frame counts. scene.audio is relative to public/, for example audio/scene-01.wav. For ffmpeg, likewise derive each visual segment's length from the successful audio duration and any explicit pause. Listen to transitions and check the final frame for clipped words.

Natiq has no word timestamps. Prefer scene-level captions, or label proportional caption timings as estimated and manually review them. Use a separate forced alignment tool if the user needs accurate word highlighting. Render a preview, check Arabic shaping/RTL text, voice/music balance, pacing and pronunciation, then render the requested output. Refer to the README example for a minimal Remotion composition and render command.