ComfyUI Extension: ComfyUI Zonos TTS Node

Authored by BahaC

Created

Updated

26 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.

A ComfyUI custom node that brings Zonos Text-to-Speech capabilities to your workflows, featuring high-quality speech synthesis and voice cloning.

Looking for a different extension?

Custom Nodes (1)

README

ComfyUI Zonos TTS Node

A ComfyUI custom node that brings Zonos Text-to-Speech capabilities to your workflows, featuring high-quality speech synthesis and voice cloning.

Features

  • 🎯 High-quality text-to-speech synthesis
  • 🗣️ Voice cloning from reference audio
  • 💾 Local model caching for faster loading
  • 🎚️ Advanced parameter control for speech generation
  • 🌍 Support for English, Japanese and many other languages.
  • ⚡ Multiple model architectures (Transformer/Hybrid)

Installation

  1. Clone this repository into your ComfyUI custom nodes directory:
cd ComfyUI/custom_nodes/
git clone https://github.com/BahaC/ComfyUI-ZonosTTS.git
  1. Install the requirements:
cd ComfyUI-ZonosTTS
pip install -r requirements.txt

Node Usage

Zonos Text to Speech

The node provides a simple interface for text-to-speech conversion with advanced options:

Inputs

  • text: Input text to synthesize (String)
  • language: Language code selection (en-us, ja-jp)
  • model_name: Choice of model architecture:
    • Zyphra/Zonos-v0.1-transformer: Faster, lighter model
    • Zyphra/Zonos-v0.1-hybrid: Higher quality (requires additional dependencies)
  • audio_file: Reference audio for voice cloning (optional)
  • cfg_scale: Control over generation quality (1.0 - 10.0)

Output

  • audio_path: Path to the generated WAV file

Model Management

Models are automatically downloaded and cached in:

/workspace/ComfyUI/models/TTS/Zonos/

The node implements smart model caching:

  • First run: Downloads and caches the model
  • Subsequent runs: Uses cached model for faster loading
  • Automatic model switching when changing architectures

Example Workflows

Basic Text to Speech

[Text Input] -> [Zonos TTS] -> [Audio Output]

Voice Cloning

[Text Input] -> [Zonos TTS] <- [Audio File] == [Audio File]

Configuration

Audio Output

Generated audio files are saved with unique timestamps:

output/zonos_YYYYMMDD-HHMMSS_UUID.wav

Model Settings

  • Transformer Model

    • Faster inference
    • Lower resource requirements
    • Good for most use cases
  • Hybrid Model

    • Higher quality output
    • Requires additional dependencies
    • More resource intensive

Requirements

  • Python >= 3.10
  • torch >= 2.0.0
  • torchaudio >= 2.0.0
  • safetensors >= 0.3.0
  • huggingface_hub >= 0.16.0
  • Additional dependencies in requirements.txt

Troubleshooting

Common Issues

  1. Model Download Fails

    • Check your internet connection
    • Ensure you have sufficient disk space
    • Try manually downloading to the models directory
  2. Voice Cloning Issues

    • Ensure reference audio is clean and contains only speech
    • Use WAV format for reference audio
    • Keep reference audio under 30 seconds
  3. CUDA Out of Memory

    • Try using the transformer model instead of hybrid
    • Reduce batch size or audio length
    • Free up GPU memory from other applications

Credits

License

This project is licensed under the terms of the LICENSE file included in the repository.

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