Audio Generation & Import Guide
Generate voices and SFX natively in Story mode, or bring your own audio
🎬 Story Mode Audio (Recommended) v3.0.0+
As of v3.0.0 you no longer need to call external APIs yourself. The ⑤ Audio step of Story mode generates dialogue voices (TTS) and sound effects (SFX) right inside the app.
Create segments with ④ Scene Split
Splitting the script into scenes breaks each scene into narration/dialogue segments and sfx segments. Every segment carries a speaker and an emotion.
Assign a voice per speaker in ⑤ Audio
Click the 🎙 button next to a speaker to open the voice picker. Speakers you leave alone are generated with the default voice.
Run it, then check segment by segment
Generated audio is written to story/audio/segments/ in your project, and once the step finishes a per-speaker track timeline appears in the panel. Each row of the table offers ▶Test · Preview · ↻Regenerate.
💡 Tip: Tick "Auto" on the steps you trust and hit "Run all" — scene split → audio → prompts then run back to back. For the full Story-mode walkthrough, see the Story Mode Guide.
🎙️ Per-Speaker Voice Assignment
Voices are assigned per speaker. Within one script you can mix engines freely — Typecast for the narrator, ElevenLabs for the lead, Gemini for a supporting character.
Three TTS Engines
| Engine | Description |
|---|---|
Typecast | Add your API key in Settings → TTS keys and the voice list loads. Supports the emotion parameter directly. |
Gemini TTS | Reuses your Gemini API key — no separate key needed. Offers multilingual prebuilt voices. |
ElevenLabs | Add your API key in Settings → TTS keys. There are far too many shared voices to preload, so typing 2+ characters in the picker search fetches more remotely. |
The Voice Picker
Good to know
— If the character's gender differs from the voice's gender, a ⚠ appears on the button. It is only a warning — generation still runs.
— Voice cards whose gender is not confirmed can be right-clicked to mark them female or male yourself.
— Every speaker gets its own narration track (the narrator is track 0), so the export lands in CapCut as separate tracks.
🎭 How Emotion Is Applied
Emotion is attached to each segment automatically during scene split, and you can read it in the Segment (emotion) column of the ⑤ Audio table.
Emotion Values
| Emotion | Description |
|---|---|
normal | Default tone |
happy | Bright / joyful |
sad | Sad / melancholic |
angry | Angry / intense |
Rules
— Emotion applies to character lines only. Narrator segments are always synthesized as normal, whatever the emotion value says.
— Each engine expresses it differently — Typecast takes an emotion parameter, Gemini TTS uses a style instruction, ElevenLabs approximates the tone via voice settings.
— Change a speaker's voice, engine, or emotion and the affected segments are re-synthesized instead of reused.
🔊 Automatic SFX Extraction
Sound effects are pulled out of the script for you. During scene split the AI inserts an sfx segment (a short English description) wherever the story needs one — a door creaking open, distant thunder, footsteps — and the ⑤ Audio step generates it in place.
An sfx segment holds the same position in the sequence as a line of dialogue, so it lands exactly where it belongs inside the scene.
The speaker column shows SFX; the segment column shows the English description plus a source dropdown. ElevenLabs is the default and the only source that generates — Library is not available yet.
Use ▶Test to generate just that one effect and listen to it; hit ↻ to regenerate if it is not right.
Sound effects are written alongside the dialogue audio in story/audio/segments/.
🎧 Audio Tab in App
Once imported, the Audio tab provides a full overview of your audio package.
Voice count per character, SFX count per category, total duration breakdown.
Chronological list with automatic scene matching via timecodes.
Preview any audio file directly in the app with inline player.
Flag unsuitable audio for replacement. Sort by name, duration, or status.
💡 Three Results-Panel Views: The results panel below the generation tabs switches between ▶ Timeline · ☰ Results · ⊞ Grid. The Grid view was added in v3.0.0.
📦 Bring Your Own Audio Manual
If you would rather use audio you produced elsewhere, the original audio package import still works exactly as before. Use it when you are not working in Story mode, or when you already have voices and effects on hand.
Generate TTS audio externally
Use Typecast or ElevenLabs to generate per-character dialogue voices, and organize the files by character folder.
Generate SFX sound effects
Use the ElevenLabs Sound Generation API to create ambient sounds, footsteps, props, and other effects by category.
Import the package in AutoFlowCut
Place the audio files in the media/ folder of your project directory. AutoFlowCut detects and imports the audio package automatically.
💡 Tip: Audio files with timecodes in their filenames are automatically matched to the corresponding SRT subtitle timestamps.
📁 Audio Package Structure Manual
Organize your audio files in the following directory structure inside your project folder:
media/ ├── voices/ # Per-character TTS files │ ├── narrator/ │ │ ├── narrator_001_0000.mp3 │ │ └── narrator_002_0035.mp3 │ └── scholar/ │ └── scholar_001_0120.mp3 ├── sfx/ # Sound effects by category │ ├── 01_props/ │ ├── 02_ambience_wind/ │ ├── 03_breath/ │ ├── 04_footsteps/ │ ├── 05_metal_doors/ │ ├── 06_writing/ │ └── 07_crowd/ └── .audio_review.json # Tracks unsuitable audio
💡 Note: The voices/ folder contains per-character subfolders. The sfx/ folder contains category-based subfolders.
🎙️ TTS (Dialogue Voice) Manual
Per-character dialogue audio generated via Text-to-Speech APIs.
Generation API: Typecast
API: https://api.typecast.ai/v1/text-to-speech
File Naming Convention
{character}_{number}_{MMSS}.mp3
Examples:
narrator/narrator_001_0000.mp3 # narrator, line 1, at 00:00
scholar/scholar_003_0245.mp3 # scholar, line 3, at 02:45
The timecode (MMSS) enables automatic matching with SRT subtitle timestamps.
Emotion Parameters
The emotion values are the same ones the app uses (normal / happy / sad / angry). See How Emotion Is Applied for the details.
🔊 SFX (Sound Effects) Manual
Ambient sounds, foley, and effects generated via AI sound generation.
Generation API: ElevenLabs
API: https://api.elevenlabs.io/v1/sound-generation
7 SFX Categories
01_props/ — Object interactions, clicks02_ambience_*/ — Wind, rain, birds03_breath/ — Breathing, sighing04_footsteps/ — Walking, running05_metal_doors/ — Doors, locks, impacts06_writing/ — Brush, pen strokes07_crowd/ — Murmuring, chatterFile Naming
{category}/{descriptive_name}.mp3
Timecoded SFX (synced to a specific scene):
abacus_beads_dark_01_0015.mp3 # for the 00:15 scene
abacus_beads_01_0134.mp3 # for the 01:34 scene
Timecoded filenames (ending in _MMSS) are automatically matched to corresponding scenes.
🔍 Audio Review System Manual
Flag unsuitable audio files through the app UI or via Claude Code MCP tools. Flagged files are tracked in .audio_review.json.
.audio_review.json Structure
{
"media/sfx/02_ambience_wind/wind_howl_01.mp3": {
"status": "flagged",
"reason": "No timecode",
"flaggedAt": "2026-03-16T05:44:55.228Z"
}
}
MCP Tools
list_audio_reviews
— View list of flagged audio files
update_audio_review
— Add or remove flags