ComfyUI Extension: DVA_Qwen_TTS

Authored by SLVGITHUB

Created

Updated

1 stars

Run ComfyUI workflows without the setup

No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

High-quality multilingual TTS with voice cloning, emotion control, and support for Qwen3-TTS models (CustomVoice, VoiceDesign, Base).

Looking for a different extension?

Custom Nodes (0)

    README

    <div align="center">

    🎤 Qwen-TTS Nodes for ComfyUI

    Qwen-TTS Banner ComfyUI Compatible License

    Nodes to integrate Qwen3-TTS into ComfyUI with emotion support and voice cloning

    Python Version PyTorch Hugging Face

    </div>

    📋 Summary

    ✨ Features

    <div align="center"> <img src="https://raw.githubusercontent.com/SLVGITHUB/QWEN3_TTS_DVA/main/images/features.png" alt="Qwen-TTS" width="800"/> </div>
    • 🎭 Support of emotions - synthesis of speech with different emotional colors
    • 🎤 Voice cloning - creation of voice duets from reference audio
    • 🌍 Multilingual - support for Russian, English and other languages
    • ⚡ High performance - optimization for CUDA and CPU
    • 🎨 Flexible setting - fine adjustment of synthesis parameters
    • 🔄 Batch processing - mass audio file generation

    🚀 Installation

    Method 1: Through ComfyUI Manager (recommended)

    1. Open ComfyUI Manager
    2. Go to Custom Nodes Install Install via Git URL
    3. Enter the URL: https://github.com/SLVGITHUB/QWEN3_TTS_DVA
    4. Press Install
    5. Restart the ComfyUI

    Method 2: Manual Installation

    `bash

    Clone repository to custom_nodes directory

    cd ComfyUI/custom_nodes git clone https://github.com/SLVGITHUB/QWEN3_TTS_DVA.git

    Set dependencies

    pip install -r requirements.txt

    Or set manually

    pip install qwen-tts soundfile openai-whisper faster_whisper `

    Requirements

    • Python 3.8+
    • ComfyUI latest version
    • PyTorch 2.0+
    • Video card with CUDA support (recommended) or CPU

    🎯 Nodes

    📦 Qwen TTS Loader

    Downloads Qwen-TTS speech model to memory

    <div align="center"> <img src="https://raw.githubusercontent.com/SLVGITHUB/QWEN3_TTS_DVA/main/images/model_loader.png" alt="Model Loader" width="400"/> </div>

    Supported models:

    • «Qwen3-TTS-Base`- for voice cloning
    • «Qwen3-TTS-CustomVoice» - a synthesis called speaker
    • «Qwen3-TTS-VoiceDesign» - text description synthesis

    Parameters:

    • Calculation accuracy: fp16, bf16, fp32
    • Device: CUDA, CPU -Type of attention: standard, optimized

    🎤 Qwen TTS Generate

    Generates speech from text without reference audio

    <div align="center"> <img src="https://raw.githubusercontent.com/SLVGITHUB/QWEN3_TTS_DVA/main/images/text_to_speech.png" alt="Speech Generation" width="400"/> </div>

    Features:

    • Support languages: Russian, English, Chinese, Japanese and others
    • Emotional presets: neutral, cheerful, sad, angry, scared
    • Extended parameters: temperature, top-p, sample length
    • For CustomVoice: specifying the speaker’s name (e.g., "Vivian", "Alex", "Maya")

    🎭 Qwen TTS Voice Clone

    Clones voice from reference audio file

    <div align="center"> <img src="https://raw.githubusercontent.com/SLVGITHUB/QWEN3_TTS_DVA/main/images/voice_cloning.png" alt="Voice cloning" width="400"/> </div>

    Requirements:

    • Input audio (reference) - WAV, MP3, FLAC
    • Audio text (ref_text)
    • New text for synthesis

    Perfect for:

    • Voice doubling
    • Voice-overs of content with a unique voice
    • Remaking of historical speeches

    Attention!!! Reference audio must be in the following format: mono, 44.1kHz, PCM, 16bit

    📚 Qwen TTS Batch Generate

    Generates multiple audio files per run

    <div align="center"> <img src="https://raw.githubusercontent.com/SLVGITHUB/QWEN3_TTS_DVA/main/images/batch_generation.png" alt="Batch generation" width="400"/> </div>

    Functional:

    • Split text by specified separator (default is "|")
    • Parallel or sequential processing
    • Automatic file numbering
    • Support of different parameters for each segment

    💾 Qwen TTS Audio Saver

    Saves generated audio to disk

    <div align="center"> <img src="https://raw.githubusercontent.com/SLVGITHUB/QWEN3_TTS_DVA/main/images/audio_saver.png" alt="Save audio" width="400"/> </div>

    Save settings:

    • Format: WAV (16-bit, 24kHz)
    • Destination folder: «ComfyUI/output/tts/
    • Automatic file deletion
    • Metadata in JSON format
    • Overwrite or incremental save

    🔀 Qwen TTS Emotion Mixer

    Mixes options with different emotions

    <div align="center"> <img src="https://raw.githubusercontent.com/SLVGITHUB/QWEN3_TTS_DVA/main/images/emotion_mixer.png" alt="Mixing emotions" width="400"/> </div>

    Application:

    • Creating complex emotional transitions
    • Mixing 70% "calm" + 30% "energetic"
    • Real-time adjustment of weights
    • Normalization of the sum of weights up to 1.0

    🎨 Example Workflow

    <div align="center"> <img src="https://raw.githubusercontent.com/SLVGITHUB/QWEN3_TTS_DVA/main/images/workflow.png" alt="workflow" width="400"/> </div>

    Typical use case:

    1. Upload the model via Qwen TTS Loader
    2. Generate speech via Qwen TTS Generate
    3. Set the save settings to Qwen TTS Audio Saver
    4. Start workflow

    🔧 Model Parameters

    Recommended settings

    | Parameter | Qwen3-TTS-Base | Qwen3-TTS-CustomVoice | Qwen3-TTS-VoiceDesign | |--------|______|| | Temperature | 0.6-0.8 | 0.7-0.9 | 0.7-0.9 | | Top-P | 0.8-0.95 | 0.85-0.98 | 0.85-0.98 | | Sample length | 2048 | 1024 | 1024 |

    Supported languages

    • 🇷🇺 Russian (ru)
    • 🇺🇸 English (en)
    • 🇨🇳 Chinese (zh)
    • 🇯🇵 Japanese (ja)
    • 🇰🇷 Korean (ko)
    • 🇫🇷 French (fr)
    • 🇩🇪 German (de)
    • 🇪🇸 Spanish (es)

    📁 Project structure

    QWEN3_TTS_DVA/ qwen_tts_comfy/ nodes.py # Basic ComfyUI Nodes __init__.py requirements.txt # Python dependencies README.md examples/ workflows/ audio_samples/ images/ # Images for documentation License

    ❓ Frequently asked questions

    ❓ Which model to choose?

    • For voice cloning: Qwen3-TTS-Base
    • For ready voices: Qwen3-TTS-CustomVoice
    • To create unique voices: Qwen3-TTS-VoiceDesign

    ❓ Why does it work slowly on the CPU?

    TTS models require significant computing resources. It is recommended to use a GPU with CUDA support.

    ❓ How to improve synthesis quality?

    • Use longer reference audio for cloning
    • Experiment with temperature parameters and top-p
    • Use emotional presets for expressiveness

    ❓ Are other audio formats supported?

    Input audio: WAV, MP3, FLAC, OGG Audio output: WAV (standard), can be converted through additional nodes

    📄 License

    This project is distributed under the Apache 2.0. license.

    🔗 Useful links

    <div align="center">

    🌐 Official Qwen3-TTS Repository | 🤗 Models on Hugging Face | 💬 Problem Discussion

    </div>

    🤝 Contribution to the project

    Welcome:

    • Error messages
    • Proposals for improvement
    • Pull requests
    • Sample workflows
    <div align="center">

    Created with ❤️ for ComfyUI community

    ⭐ If you like this project, put a star on GitHub!

    </div>

    Run ComfyUI workflows without the setup

    No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

    Learn more