🌋 ComfyUI LavaSR

A custom node for ComfyUI that integrates the lightweight LavaSR speech enhancement model. LavaSR enhances low-quality audio with speeds reaching roughly 5000x realtime on GPU and over 60x realtime on CPU.

Features

• Extremely Fast: Achieves enhancement speeds of up to 5000x real-time on modern GPUs and 50x real-time on CPUs.
• High-Quality Speech Enhancement: Converts low-quality or noisy input audio into crisp, studio-quality 48kHz audio.
• High Efficiency: Extremely lightweight, requiring just ~500MB VRAM to run.
• Local Model Support: Automatically detects and loads models placed in ComfyUI/models/lavasr/. Falls back to automatically downloading the default model (YatharthS/LavaSR) from Hugging Face if none are found locally.
• Denoising & Batching: Provides dedicated toggles for heavy background denoising and batch processing (crucial for preventing OOM on very long audio files).

💡 Tip: This node pairs exceptionally well with ComfyUI-KittenTTS for fast, low-VRAM voice generation. You can pipe the output from KittenTTS directly into LavaSR for incredibly crisp results.

Installation

Method 1: ComfyUI Manager (Recommended)

You can install this node via the ComfyUI Manager by searching for "LavaSR" or by installing from this Git URL. The dependencies should install automatically.

Method 2: Manual Installation

This extension requires the LavaSR python package to be installed in your ComfyUI python environment.

1. Navigate to your ComfyUI custom_nodes directory:

cd ComfyUI/custom_nodes

2. Clone this repository:

git clone https://github.com/NightMean/ComfyUI-LavaSR.git

3. Install the required LavaSR python package.

For standard Python environments:

python -m pip install git+https://github.com/ysharma3501/LavaSR.git

For ComfyUI Portable (Windows Standalone): Open a terminal in your main ComfyUI folder and run:

.\python_embeded\python.exe -m pip install git+https://github.com/ysharma3501/LavaSR.git

Model Setup

By default, the node will automatically download the required model from Hugging Face into ComfyUI/models/lavasr/YatharthS_LavaSR upon first use.

Method 3: Manual Model Download (Optional)

If you prefer to download the model manually, or if your environment blocks automatic Hugging Face downloads:

1. Download the files from YatharthS/LavaSR on Hugging Face.
2. Place the downloaded files or folder into: ComfyUI/models/lavasr/YatharthS_LavaSR
3. Restart ComfyUI and select your custom folder name from the model_id dropdown.

Usage

You can find an example workflow in example_workflow/LavaSR_example_workflow.json

1. Connect a Load Audio node.
2. Connect it to the audio input of LavaSR Enhance.
3. Add a Save Audio node and connect the audio output from the LavaSR node to it.

Node Parameters

• model_id: Select the HuggingFace ID YatharthS/LavaSR (loads automatically via internet) or select a locally cached directory name if you placed the models manually inside ComfyUI/models/lavasr/.
• version: Select LavaEnhance2 (default, highest quality) or LavaEnhance1.
• sampling_rate: The assumed source quality of your input audio. Best practice is to match the source audio's sample rate (e.g., 8000Hz for old phones, 16000Hz for standard mic audio).
• denoise: Set to True to heavily denoise the background audio before running enhancement.
• batch_processing: Set to True to process highly sustained, long audio files (prevents Out-Of-Memory errors).

Donations

To support me you can use link below:

Author & Credits

• LavaSR by @ysharma3501
• ComfyUI node by @NightMean

License

Apache 2.0 - See LICENSE for details.