billwuhao-ComfyUI_CSM/README.md

中文|English

ComfyUI Node for CSM

CSM (Conversational Speech Model), a model that supports multi-person conversations, voice cloning, and generates speech with corresponding emotional changes based on the emotional changes in the conversation. Unfortunately, it is currently only available in English. This node temporarily supports simultaneous conversations with up to 10 people.

Recording nodes can be interspersed within to create multi-person conversations.

It also supports audio watermark detection (automatic watermark detection) and audio adding encrypted watermarks.

📣 Updates

[2025-03-18]⚒️: Released version v1.0.0.

Detailed node usage example workflows: example_workflows

Installation

cd ComfyUI/custom_nodes
git clone https://github.com/billwuhao/ComfyUI_CSM.git
cd ComfyUI_CSM
pip install -r requirements.txt

# python_embeded
./python_embeded/python.exe -m pip install -r requirements.txt

Model Download

• csm-1b: Download config.json and model.safetensors and place them in the ComfyUI/models/TTS/csm-1b directory.

• moshiko-pytorch-bf16: Download tokenizer-e351c8d8-checkpoint125.safetensors and place it in the ComfyUI/models/TTS/moshiko-pytorch-bf16 directory.

• SilentCipher: Download all models and place them in the ComfyUI\models\TTS\SilentCipher\44_1_khz\73999_iteration directory.

• Llama-3.2-1B: Download everything except the original directory and place it in the ComfyUI\models\LLM\Llama-3.2-1B directory.

Acknowledgements

csm

Thanks to the SesameAILabs team for their excellent work 👍.