ComfyUI-kaola-moss-tts

A ComfyUI custom node pack for the MOSS-TTS Family — a suite of open-source speech and audio generation models. This pack provides nodes for dialogue synthesis, narration, voice design, and sound effects.

Features

• 🎙️ Multi-Speaker Dialogue (MOSS-TTSD): Generate expressive multi-party conversations with up to 5 speakers using [S1]~[S5] tags. Supports long-form (up to 60 min) coherent audio.
• 🗣️ Narration & TTS (MOSS-TTS): High-fidelity single-speaker speech synthesis with zero-shot voice cloning. Available in 8B and 1.7B variants.
• 🎨 Voice Design (MOSS-VoiceGenerator): Create speaker timbres from text descriptions — no reference audio needed.
• 🔊 Sound Effects (MOSS-SoundEffect): Generate sound effects from text prompts with controllable duration.
• 🔄 Zero-Shot Voice Cloning: Clone any voice with just a short reference audio clip.
• 🌍 Multilingual: Supports 20+ languages including Chinese, English, Japanese, and more.

Installation

[!TIP] We strongly recommend creating a fresh Python 3.12 environment to avoid dependency conflicts:

conda create -n moss_ttsd python=3.12 -y && conda activate moss_ttsd
pip install flash-attn --no-build-isolation

1. Clone this repository into your ComfyUI/custom_nodes directory:

   cd custom_nodes
   git clone https://github.com/kana112233/ComfyUI-kaola-moss-tts.git
   

2. Install dependencies:

   cd ComfyUI-kaola-moss-tts
   pip install -r requirements.txt
   

   Note: For GPU acceleration, ensure you have a CUDA-compatible PyTorch installed.

[!WARNING] This node requires transformers >= 5.0.0. Please make sure your environment meets this requirement:

pip install "transformers>=5.0.0"

Nodes Overview

Models

Auto Download

If models are not found locally, they will be automatically downloaded from HuggingFace on first run.

Manual Download

Place models in the ComfyUI/models/moss_ttsd/ directory:

ComfyUI/
└── models/
    └── moss_ttsd/
        ├── MOSS-TTSD-v1.0/
        ├── MOSS-TTS/
        ├── MOSS-VoiceGenerator/
        ├── MOSS-SoundEffect/
        └── MOSS-Audio-Tokenizer/

You can use huggingface-cli to download:

huggingface-cli download OpenMOSS-Team/MOSS-TTSD-v1.0 --local-dir ComfyUI/models/moss_ttsd/MOSS-TTSD-v1.0
huggingface-cli download OpenMOSS-Team/MOSS-TTS --local-dir ComfyUI/models/moss_ttsd/MOSS-TTS
huggingface-cli download OpenMOSS-Team/MOSS-TTS-Local-Transformer --local-dir ComfyUI/models/moss_ttsd/MOSS-TTS-Local-Transformer
huggingface-cli download OpenMOSS-Team/MOSS-VoiceGenerator --local-dir ComfyUI/models/moss_ttsd/MOSS-VoiceGenerator
huggingface-cli download OpenMOSS-Team/MOSS-SoundEffect --local-dir ComfyUI/models/moss_ttsd/MOSS-SoundEffect
huggingface-cli download OpenMOSS-Team/MOSS-Audio-Tokenizer --local-dir ComfyUI/models/moss_ttsd/MOSS-Audio-Tokenizer

Usage

1. MOSS-TTSD — Multi-Speaker Dialogue

Best for podcasts, audiobooks, and conversational content.

• Load Model: Load MOSS-TTSD Model node
• Load Codec: Load MOSS-Audio Codec node
• Generate: MOSS-TTSD Generate node
• Text format: Use [S1], [S2]...[S5] to switch speakers

  [S1] Hello there! [S2] Hi! How are you doing today?
  

Generation Modes

2. MOSS-TTS — Narration & Single-Speaker TTS

Best for voiceovers, dubbing, and long-form narration.

• Load Model: Load MOSS-TTS Foundation Model node (choose 8B or 1.7B)
• Generate: MOSS-TTS Generate node
  • Text: Narration content
  • Reference Audio: (Optional) Connect for zero-shot voice cloning
  • Instruction: (Optional) Describe the desired voice style

3. MOSS-VoiceGenerator — Voice Design

Design speaker timbres from text descriptions — no reference audio needed!

• Load Model: Load MOSS Voice Generator Model node
• Generate: MOSS Voice Generator Generate node
  • Instruction: Describe the voice (e.g., "A warm, deep male voice", "年轻女性，温柔的声音")
  • Text: Content to speak with this designed voice
  • Output: Can be used as reference audio for MOSS-TTSD voice_clone mode!

4. MOSS-SoundEffect — Sound Effect Generation

Generate environmental sounds, ambient audio, and sound effects from text prompts.

• Load Model: Load MOSS Sound Effect Model node
• Generate: MOSS Sound Effect Generate node
  • Text: Description (e.g., "birds chirping in a forest", "rain on a tin roof")
  • Duration: Target duration in seconds (~12.5 tokens/sec)

Example Workflows

Example workflow files are provided in the examples/ directory:

License

Apache 2.0 (Inherited from MOSS-TTS)