ComfyUI-GLM-TTS

ComfyUI nodes for GLM-TTS, a high-quality text-to-speech system supporting zero-shot voice cloning.

Installation

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

   cd ComfyUI/custom_nodes
   git clone https://github.com/karas17/ComfyUI-GLM-TTS.git
   

   (If you downloaded this folder directly, just rename it to ComfyUI-GLM-TTS)

2. Install dependencies:

   pip install -r requirements.txt
   

   Note:

   • torch, torchaudio, and transformers are required but expected to be in your ComfyUI environment.
   • onnxruntime (or onnxruntime-gpu) is required for the frontend. If you encounter issues, install onnxruntime-gpu.

3. Models: By default, models are loaded from ComfyUI/models/GLM-TTS. If missing, they will be auto-downloaded from HuggingFace zai-org/GLM-TTS to that location. Windows 示例: <ComfyUI 安装目录>\models\GLM-TTS

   Structure expected inside the model path:

   • speech_tokenizer/
   • vq32k-phoneme-tokenizer/
   • llm/
   • flow/
   • frontend/ (contains campplus.onnx, spk2info.pt)

   You can also provide an absolute model_path to a prepared directory with the above structure.

Usage

1. GLM-TTS Loader:
   • model_path: Path to your models. Default GLM-TTS resolves to ComfyUI/models/GLM-TTS. Absolute paths are supported（Windows 示例：<ComfyUI 安装目录>\models\GLM-TTS）.
   • sample_rate: 24000 (default) or 32000.
2. GLM-TTS Sampler:
   • model: Connect from Loader.
   • text: Text to speak.
   • reference_audio (Optional): Audio for voice cloning.
   • reference_text (Optional): Transcript of the reference audio (improves cloning).

Notes

• The first run might be slow as it loads models.
• Ensure your reference_audio is clear.