VoxFlow AI is a desktop application designed to streamline the creation of audiobooks and audio content. Built with modern web and Rust technologies, it allows you to visually script, synthesize natural-sounding speech using Alibaba Cloud Bailian, and manage your audio projects with ease.
Note: This application uses the Alibaba Cloud Bailian (DashScope) platform for its Text-to-Speech (TTS) capabilities. You will need a valid API Key from the platform to use the synthesis features.
- 🎙️ Alibaba Cloud Bailian Integration — High-quality TTS synthesis powered by models like Qwen3 TTS Flash.
- 🎬 AI Video Generation — Automatically generate visually rich video accompaniments for your audiobook. The AI creates atmospheric motion graphics — geometric animations, gradient flows, kinetic typography — that complement the narration without cluttering the screen with text. Each section is rendered independently with audio-synced timing, then merged into a final video. Supports sleep mode audio processing for bedtime listening.
- ⚙️ User-Configurable Settings — Easily input your API Key, select models, and adjust default voice settings (speed, pitch, voice) directly from the application UI.
- ⏱️ Adjustable Intervals — Fine-tune the silence between script lines to ensure natural pacing.
- 💾 Robust Data Management — Built-in SQLite database ensures data integrity and prevents conflicts during manual script creation.
- 🖥️ Modern Desktop Experience — Powered by Tauri 2.0 for a lightweight, secure, and native feel.
Follow these instructions to get a local copy of the project up and running.
- Node.js (v18 or later)
- Rust (latest stable version)
- FFmpeg (for audio/video processing)
- Hyperframes (
npm install -g hyperframes— used for HTML-to-video rendering)
-
Clone the repository:
git clone https://github.com/iMyth/VoxFlow.git cd VoxFlow -
Install dependencies:
# Install frontend dependencies pnpm install # Or: npm install
-
Configure Bailian API:
Unlike many CLI tools, VoxFlow does not require you to set environment variables manually.
- Launch the application (see Development below).
- Click the Settings (gear icon ⚙️).
- Enter your DashScope API Key in the designated field.
To start the application in development mode with hot-reloading:
npm run tauri:devThis command will start both the Vite development server and the Tauri backend simultaneously.
VoxFlow generates visual accompaniment videos for each audiobook section:
- Audio merge — Section audio fragments are concatenated with silence gaps, then measured via ffprobe to get the exact duration.
- LLM composition — An AI agent generates an HTML/GSAP animation composition (via Hyperframes format) tailored to the content's mood and themes. The prompt emphasizes visual richness (6-10 elements/scene) while avoiding text-heavy subtitle-style output.
- Post-processing — The generated HTML undergoes multiple safety passes:
- Force-correct
data-durationon the composition element (ignoring CSS selector false positives) - Force-set
window.__hf.durationto the ffprobe-measured audio duration - Clamp any clip extending beyond the audio duration
- Fix CSS font variables and ensure timeline registration
- Force-correct
- Render —
npx hyperframes renderconverts the HTML to a silent MP4 (headless Chrome capture at 30fps). - Audio merge — FFmpeg merges the silent video with section audio using
-shortestto guarantee the output matches audio length. - Final merge — All section videos are concatenated (with resolution normalization to 1920×1080 if needed) into the final output.
On first run, macOS will prompt for App Management permission (required by headless Chrome). VoxFlow handles this automatically — it detects the permission failure, shows a toast notification, waits for you to grant access, and retries.
To build the application for your platform:
npm run tauri:build| Layer | Technologies |
|---|---|
| Frontend | React 19, TypeScript, Vite, Tailwind CSS, Zustand |
| Backend | Rust, Tauri 2.0, Tokio, Rusqlite (SQLite) |
| AI Services | Alibaba Cloud Bailian (DashScope), Qwen3-Plus (LLM), Hyperframes (HTML→Video) |
This project is licensed under the MIT License.