Add Voiceover — Generate & Integrate Narration

You are helping the user add a voiceover narration to their Remotion project.

Load the remotion-production skill for the voiceover-sync and audio-integration rules, and the official ElevenLabs text-to-speech skill (npx skills add elevenlabs/skills) for the API details.

Workflow

0. Preview First

If Studio isn't running yet, start it now with npx remotion studio (or --no-open with an in-app browser, then open the URL it prints) and keep it open, following the official remotion-best-practices skill.

1. Get the Script

If the user hasn't provided a script:

• Ask what the video is about
• Ask about tone (professional, casual, energetic, calm)
• Write the narration script, split into one line or paragraph per scene
• Present for approval

2. Choose Voice

Pre-made voices (from the official text-to-speech skill):

• George (JBFqnCBsd6RMkjVDRZzb) — Male, narrative (default)
• Sarah (EXAVITQu4vr4xnSDxMaL) — Female, soft
• Daniel (onwK4e9ZLuTAKqWW03F9) — Male, authoritative
• Charlotte (XB0fDUnXU5powFXDhCwa) — Female, conversational

For a specific mood, accent or language, search the voice library; for a cloned voice or several speakers, see the elevenlabs-advanced rule. Use the same voice for every scene.

3. Generate TTS, One File per Scene

Generate through ElevenLabs, as the official Remotion voiceover guide also recommends:

1. If ELEVENLABS_API_KEY isn't set for the project, run npx -y @elevenlabs/cli@latest onboard init from the project root and follow what it prints. Never ask for the key in chat.
2. npm i @elevenlabs/elevenlabs-js, then create scripts/generate-audio.ts from the audio-integration rule, with one voiceover entry per scene (vo-scene1, vo-scene2, …) and the chosen VOICE_ID. The model is eleven_v4.
3. Run node scripts/generate-audio.ts from the project root. Each scene lands at public/vo-sceneN.mp3. To redo one line, delete its file and run the script again.

The script reports no duration; step 4 measures the files.

Alternatives:

• The ElevenLabs MCP text_to_speech tool is deprecated (archived upstream) but still ships for quick interactive one-offs. It writes into public/; rename its output to vo-sceneN.mp3.
• KIE generate_speech (remotion-media) for cheap drafts: text, output_name: vo-scene1, voice: Eric (21 named voices, listed in the voiceover-sync rule). Suno and the other KIE-resold models aren't licensed by their owners through KIE. Use KIE audio for drafts and internal work, and generate final commercial audio with ElevenLabs.

4. Size the Scenes with calculateMetadata

Measure each vo-sceneN.mp3 in calculateMetadata with Mediabunny and pass the per-scene frame counts (Math.ceil(seconds * fps)) as props, as the voiceover-sync rule shows. The composition length then follows the audio, and regenerating a line updates the timing in Studio.

5. Wire into Composition

Place each scene's file with <Audio> from @remotion/media inside that scene (install with npx remotion add @remotion/media), following the official remotion-markup skill (audio.md):

import { Audio } from "@remotion/media";
import { staticFile } from "remotion";

// Inside scene 1:
<Audio src={staticFile("vo-scene1.mp3")} volume={1} />

6. Captions (optional)

Offer captions. They follow the official remotion-captions skill (word-level, @remotion/whisper-webgpu), with an ElevenLabs Scribe fallback when there's no GPU and TikTok-style presets from the captions-workflow rule. Run /add-captions, or /transcribe if the user only wants an .srt file.

7. Duck Background Music (if present)

If the project already has background music, lower its volume during voiceover. See audio-integration rule for the ducking pattern.

8. Review in Studio

Ask the user to play it in the open Studio. Don't render unless they ask for an MP4.