ComfyUI-NeuTTS

ComfyUI custom nodes for NeuTTS — Neuphonic's on-device TTS with instant voice cloning.

---

Nodes

Node	Description
NeuTTS 🎙️	TTS + voice cloning. Takes text + optional ref audio → generates speech
NeuTTS Encode Reference 🎤	Pre-encode a reference audio once; reuse across multiple NeuTTS nodes
NeuTTS Unload Models 🗑️	Force-clear all cached models to free GPU/RAM

---

Installation

Via ComfyUI Manager

Search ComfyUI-NeuTTS → Install.

Manual

cd ComfyUI/custom_nodes
git clone https://github.com/AsifOnWeb3/ComfyUI-NeuTTS
cd ComfyUI-NeuTTS
pip install -r requirements.txt

Install NeuTTS

# CPU only
pip install neutts

# GGUF support (llama-cpp-python)
pip install neutts[llama]

# ONNX codec (faster decode)
pip install neutts[onnx]

# Everything
pip install neutts[all]

For GPU-accelerated llama-cpp-python (CUDA):

CMAKE_ARGS="-DGGML_CUDA=ON" pip install "neutts[llama]" --force-reinstall --no-cache-dir

---

Model Directory (optional local cache)

Pre-download models to avoid HuggingFace timeouts:

ComfyUI/
└── models/
    └── neutts/
        └── neuphonic/
            ├── neutts-air/
            ├── neutts-nano/
            └── neucodec/

The nodes auto-detect local models before falling back to HuggingFace Hub.

---

Available Backbones

Model	Size	Lang	Notes
neuphonic/neutts-air	~552M	EN	Best quality
neuphonic/neutts-air-q4-gguf	Q4	EN	Fastest, needs llama-cpp
neuphonic/neutts-nano	~229M	EN	Fast
neuphonic/neutts-nano-q4-gguf	Q4	EN	CPU-friendly
neuphonic/neutts-nano-french	~229M	FR
neuphonic/neutts-nano-german	~229M	DE
neuphonic/neutts-nano-spanish	~229M	ES

Codec: neuphonic/neucodec (default) or neuphonic/neucodec-onnx-decoder (faster).

---

Reference Audio Tips

• Format: .wav, mono, 16–44 kHz
• Length: 3–15 seconds
• Content: clean speech, no background noise
• Providing ref_text (transcript of the ref clip) significantly improves quality

---

Workflow Example

Load Audio → [NeuTTS Encode Reference] → ref_codes ─┐
                                                     ├→ [NeuTTS] → audio → Preview Audio
                     text ────────────────────────────┘

---

Acknowledgements

• NeuTTS by Neuphonic
• Inspired by ComfyUI-Qwen-TTS

License

Apache 2.0