Extensions/ComfyUI-KittenTTS
ComfyUI Extension

ComfyUI-KittenTTS

😻 A simple ComfyUI custom node for KittenTTS - an ultra-lightweight text-to-speech model. Works on CUDA and CPU.

By Saganaki22·Created 6 months ago·Updated 6 months ago· 10
Saganaki22/ComfyUI-KittenTTS
Nodes1
On cloudLocal install
Categoryaudio
Stars10
Updated6 months ago

Nodes (1)

😻 Kitten-TTS

A 24MB local TTS that runs on your CPU — but check the version before you run it

audio
Readme

😻 Kitten-TTS for ComfyUI

ComfyUI Python License

A simple ComfyUI custom node for KittenTTS - an ultra-lightweight text-to-speech model. Works on CUDA and CPU.

<img width="1496" height="1015" alt="Screenshot 2026-02-19 194339" src="https://github.com/user-attachments/assets/6af8a500-0a48-47da-adff-e020a5437e88" />

Demo

https://github.com/user-attachments/assets/d80120f2-c751-407e-a166-068dd1dd9e8d

Features

  • 🚀 Ultra-lightweight: Models from 19MB to 80MB
  • 💻 CPU & CUDA: Runs on any device, GPU optional
  • 🎯 Single node: All settings in one place
  • 📦 Auto-download: Models cached in ComfyUI/models/kittentts/
  • ⚡ Fast inference: Real-time speech synthesis
  • 🎤 8 voices: 4 male, 4 female

Installation

Prerequisites (Required)

Before installing the custom node, you must install espeak-ng and set up its environment variable:

  1. Install espeak-ng: Download and install it from https://github.com/espeak-ng/espeak-ng/.
  2. Set Environment Variable: Add a new system environment variable named PHONEMIZER_ESPEAK_LIBRARY pointing to the espeak-ng installation path (e.g., C:\Program Files\eSpeak NG\libespeak-ng.dll on Windows).
  3. Reboot: You must completely restart the ComfyUI CLI/console for the updated environment variables to be securely loaded.

Method 1: ComfyUI Manager (Recommended)

Search for "KittenTTS" in ComfyUI Manager and install.

Method 2: Manual Install

cd ComfyUI/custom_nodes
git clone https://github.com/Saganaki22/ComfyUI-KittenTTS.git

Dependencies

The node auto-installs most dependencies on first run. The correct version of onnxruntime (CPU or GPU) is dynamically chosen based on whether your PyTorch environment supports CUDA.

If you need to install or fix it manually:

Standard Python environment:

# For CPU
pip install onnxruntime

# For CUDA GPU
pip install onnxruntime-gpu

For Windows portable ComfyUI, use the embedded Python:

# From ComfyUI_windows_portable directory

# For CPU
.\python_embeded\python.exe -m pip install onnxruntime

# For CUDA GPU
.\python_embeded\python.exe -m pip install onnxruntime-gpu

Models

| Model | Params | Size | Quality | Link | |-------|--------|------|---------|------| | kitten-tts-mini | 80M | 80MB | Best | 🤗 Download | | kitten-tts-micro | 40M | 41MB | Good | 🤗 Download | | kitten-tts-nano | 15M | 56MB | Lightweight | 🤗 Download | | kitten-tts-nano-int8 | 15M | 19MB | Smallest | 🤗 Download |

⚠️ Note: Some users report issues with the int8 quantized model. We recommend the mini or micro versions for best results.

Usage

The node is simple - just one node with all settings:

Inputs

| Input | Type | Default | Description | |-------|------|---------|-------------| | model_name | Dropdown | mini-0.8 (80M) | Model size/quality | | device | Dropdown | auto | auto/cuda/cpu | | text | String | - | Text to synthesize | | voice | Dropdown | Jasper | Voice selection | | speed | Float | 1.0 | Speech speed (0.5-2.0) | | keep_loaded | Boolean | True | Keep model in memory | | output_stereo | Boolean | False | Stereo output | | clean_text | Boolean | True | Normalize text | | custom_model | String | "" | Custom HF model ID |

Voices

Male: Jasper, Bruno, Hugo, Leo
Female: Bella, Luna, Rosie, Kiki

Example

Just add the 🐱 KittenTTS node, type your text, select a voice, and connect the audio output to your pipeline.

┌─────────────────────────┐
│  🐱 KittenTTS           │ → Connect to audio nodes
├─────────────────────────┤
│ model: mini-0.8 (80M)   │
│ device: auto            │
│ text: "Hello world!"    │
│ voice: Jasper           │
│ speed: 1.0              │
│ keep_loaded: True       │
│ output_stereo: False    │
│ clean_text: True        │
│ custom_model_path ...   │
├─────────────────────────┤
└─────────────────────────┘

Model Storage

Models are downloaded to: ComfyUI/models/kittentts/<model_name>/

You can create symlinks to this folder if you want to share models between ComfyUI installations.

Troubleshooting

"CUDA out of memory"

  • Switch to a smaller model (nano or micro)
  • Set keep_loaded to False
  • Use CPU device

"onnxruntime not found"

Standard Python environment:

# For CPU
pip install onnxruntime

# For CUDA
pip install onnxruntime-gpu

For Windows portable ComfyUI, use the embedded Python:

# From ComfyUI_windows_portable directory

# For CPU
.\python_embeded\python.exe -m pip install onnxruntime

# For CUDA
.\python_embeded\python.exe -m pip install onnxruntime-gpu

"No module named 'kittentts'"

The bundled wheel should auto-install. If not:

Standard Python environment:

pip install https://github.com/KittenML/KittenTTS/releases/download/0.8/kittentts-0.8.0-py3-none-any.whl

For Windows portable ComfyUI:

.\python_embeded\python.exe -m pip install https://github.com/KittenML/KittenTTS/releases/download/0.8/kittentts-0.8.0-py3-none-any.whl

Links

Credits

License

Apache 2.0 - See LICENSE for details.