ComfyUI-SopranoTTS

ComfyUI custom nodes for Soprano TTS - a fast, lightweight text-to-speech model.

Features

• SopranoLoader: Load the TTS model once, reuse across multiple generations
• SopranoTTS: Generate speech from a single text input
• SopranoTTSBatch: Process multiple texts efficiently in batch
• SopranoTTSStream: Streaming generation (lmdeploy backend only)

Installation

Option 1: ComfyUI Manager (Recommended)

Search for "SopranoTTS" in ComfyUI Manager and click Install.

Option 2: Manual Installation

1. Navigate to your ComfyUI custom nodes directory:

   cd ComfyUI/custom_nodes
   

2. Clone this repository:

   git clone https://github.com/jo-nike/ComfyUI-SopranoTTS.git
   

3. Install dependencies:

   cd ComfyUI-SopranoTTS
   pip install -r requirements.txt
   

4. Restart ComfyUI

Usage

Basic Workflow

1. Add a Soprano TTS Loader node
2. Connect it to a Soprano TTS node
3. Enter your text
4. Connect the audio output to a SaveAudio node

Nodes

Soprano TTS Loader

Loads the SopranoTTS model. The model is cached, so subsequent runs reuse the loaded model.

Input	Type	Default	Description
backend	COMBO	transformers	Model backend: transformers (recommended for newer GPUs), lmdeploy, or auto
cache_size_mb	INT	10	LMDeploy KV cache size in MB
decoder_batch_size	INT	1	Batch size for audio decoder

Soprano TTS

Generate speech from text.

Input	Type	Default	Description
model	SOPRANO_MODEL	-	Model from SopranoLoader
text	STRING	-	Text to synthesize
temperature	FLOAT	0.3	Generation temperature (0.0-2.0)
top_p	FLOAT	0.95	Nucleus sampling parameter
repetition_penalty	FLOAT	1.2	Penalty for repeated tokens

Soprano TTS Batch

Process multiple texts (one per line) in a single batch.

Input	Type	Default	Description
texts	STRING	-	Multiple texts separated by newlines
(other params same as SopranoTTS)

Soprano TTS Stream

Streaming generation with lower latency. Only works with lmdeploy backend.

Input	Type	Default	Description
chunk_size	INT	1	Tokens per audio chunk
(other params same as SopranoTTS)

GPU Compatibility

• RTX 5080 / Blackwell GPUs: Use backend: transformers (lmdeploy doesn't support compute capability 12.0 yet)
• RTX 40xx / Ada GPUs: Either backend works
• RTX 30xx / Ampere GPUs: Either backend works

Audio Output

• Sample Rate: 32,000 Hz
• Channels: Mono
• Format: ComfyUI AUDIO type (compatible with SaveAudio, PreviewAudio, etc.)

Example Workflows

See the workflows/ folder for example workflow JSON files.

License

MIT