FL Qwen3 TTS

Advanced text-to-speech nodes for ComfyUI powered by Alibaba's Qwen3-TTS model family. Features voice cloning, voice design from text descriptions, predefined speakers, and a built-in fine-tuning UI with real-time training dashboard.

Features

• Voice Cloning - Clone any voice from 5-15 seconds of reference audio
• Voice Design - Create custom voices from natural language descriptions
• 9 Predefined Speakers - Ready-to-use voices across Chinese, English, Japanese, and Korean
• Fine-Tuning UI - Train custom voice models with a real-time dashboard (loss chart, progress, validation audio)
• 10 Languages - Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian
• Auto Transcription - Built-in Whisper integration for generating reference text
• Audio Codec - Encode and decode audio using the Qwen3-TTS 12Hz tokenizer

Nodes

Node	Description
Model Loader	Downloads and caches Qwen3-TTS models from HuggingFace
Tokenizer Loader	Loads the 12Hz audio tokenizer for encoding/decoding
Custom Voice	Generate speech using 9 predefined speakers with optional style instructions
Voice Design	Create voices from natural language descriptions (e.g. "a warm British female voice")
Voice Clone	Clone a voice from reference audio
Voice Clone Prompt	Pre-compute a voice prompt for efficient multi-generation cloning
Audio Encode	Encode audio to discrete codes using the tokenizer
Audio Decode	Decode discrete codes back to audio
Transcribe	Transcribe audio to text using Whisper (useful for ref_text)
Training UI	All-in-one fine-tuning with real-time training dashboard

Installation

ComfyUI Manager

Search for "FL Qwen3 TTS" and install.

Manual

cd ComfyUI/custom_nodes
git clone https://github.com/filliptm/ComfyUI-FL-Qwen3TTS.git
cd ComfyUI-FL-Qwen3TTS
pip install -r requirements.txt

Quick Start

Text-to-Speech

1. Add FL Qwen3 TTS Model Loader and select a model variant
2. Connect to Custom Voice, Voice Design, or Voice Clone node
3. Enter your text, configure voice settings, and generate

Voice Cloning

1. Load the Base model (Qwen3-TTS-12Hz-1.7B-Base)
2. Connect to Voice Clone with 5-15 seconds of reference audio
3. Optionally use Transcribe to generate ref_text for better results
4. Enter target text and generate

Fine-Tuning

1. Prepare a folder with audio files and matching .txt transcripts
2. Add the Training UI node and point it to your dataset folder
3. Configure training parameters (learning rate, epochs, etc.)
4. Run the workflow — monitor progress in the real-time dashboard
5. Use the output checkpoint with Custom Voice for inference

Models

Model	Type	Use Case
Qwen3-TTS-12Hz-1.7B-Base	Base	Voice cloning and fine-tuning
Qwen3-TTS-12Hz-1.7B-CustomVoice	Custom Voice	9 predefined speakers with style control
Qwen3-TTS-12Hz-1.7B-VoiceDesign	Voice Design	Create voices from text descriptions

Models download automatically on first use to ComfyUI/models/tts/Qwen3TTS/.

Predefined Speakers

Available with the CustomVoice model:

Speaker	Language	Description
Vivian	Chinese	Bright, edgy female
Serena	Chinese	Warm, gentle female
Uncle_Fu	Chinese	Seasoned male
Dylan	Chinese	Beijing dialect male
Eric	Chinese	Sichuan dialect male
Ryan	English	Dynamic male
Aiden	English	American male
Ono_Anna	Japanese	Japanese female
Sohee	Korean	Korean female

Dataset Format (Training)

audio_folder/
  sample1.wav
  sample1.txt    # contains transcript of sample1.wav
  sample2.mp3
  sample2.txt
  ...

Supported audio formats: .wav, .mp3, .flac, .ogg, .m4a

Requirements

• Python 3.9+
• 16GB RAM minimum (32GB+ recommended for training)
• NVIDIA GPU with 12GB+ VRAM recommended (CPU and Mac MPS supported for inference)

License

Apache 2.0