Extensions/ComfyUI-NovaSR
ComfyUI Extension

ComfyUI-NovaSR

Ultra-fast audio super resolution custom node for ComfyUI, powered by the NovaSR model.

By Saganaki22·Created 7 months ago·Updated 6 months ago· 35
Saganaki22/ComfyUI-NovaSR
Nodes1
On cloudLocal install
Categoryaudio
Stars35
Updated6 months ago

Nodes (1)

NovaSR

The 50KB node that makes video-model audio sound human again

audio
Readme

ComfyUI-NovaSR

Ultra-fast audio super resolution custom node for ComfyUI, powered by the NovaSR model.

<p align="center"> <a href="https://huggingface.co/drbaph/NovaSR/tree/main"> <img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-FFD21E" alt="Hugging Face Model"> </a> <a href="https://www.python.org/downloads/release/python-3100/"> <img src="https://img.shields.io/badge/python-3.10+-blue.svg" alt="Python 3.10+"> </a> <a href="https://github.com/comfyanonymous/ComfyUI"> <img src="https://img.shields.io/badge/ComfyUI-Custom%20Node-orange.svg" alt="ComfyUI"> </a> </p> <img width="1515" height="1012" alt="image" src="https://github.com/user-attachments/assets/a10cab1b-020c-4e4c-b326-b9cea0644dde" />

https://github.com/user-attachments/assets/f91bfa74-e390-4106-9f36-6df6e9a7b405

Features

  • Incredibly Fast: 3600x realtime speed on a single A100 GPU
  • Tiny Model: Just 50KB in size
  • High Quality: On par with models 5,000x larger
  • Simple to Use: Just one click to upscale audio to 48kHz
  • Auto Stereo→Mono: Automatically converts stereo audio to mono
  • Stereo Output Toggle: Option to output stereo for ComfyUI compatibility (duplicates mono)
  • Any Audio Format: Works with all formats ComfyUI supports (wav, mp3, flac, ogg, etc.)
  • Smart Resampling: Automatically resamples to required 16kHz input

Installation

Via ComfyUI Manager (Recommended)

  1. Open ComfyUI
  2. Click the Manager button → Install Custom Nodes
  3. Search for "NovaSR"
  4. Click Install on "ComfyUI-NovaSR"
  5. Restart ComfyUI

Manual Installation

  1. Clone or copy this repository into your ComfyUI custom_nodes directory:

    ComfyUI/custom_nodes/ComfyUI-NovaSR/

  2. Install dependencies:

    pip install -r ComfyUI/custom_nodes/ComfyUI-NovaSR/requirements.txt

  3. Download the NovaSR model and place it this location:

    • ComfyUI/models/NovaSR/

    Download from: https://huggingface.co/drbaph/NovaSR/tree/main

    Place one of these files:

    • pytorch_model.bin
    • NovaSR.safetensors

Usage

  1. Start ComfyUI
  2. Right-click → Add Node → Audio → NovaSR
  3. Load audio using Load Audio node (supports: wav, mp3, flac, ogg, aiff, and more)
  4. Connect audio output to the NovaSR node input
  5. The node will automatically:
    • Accept any sample rate audio input
    • Resample to 16kHz (NovaSR requirement)
    • Convert stereo to mono if needed (NovaSR is mono-only)
    • Upscale to 48kHz at 3600x realtime speed
    • Output a before/after spectrogram comparison image

Note: NovaSR outputs mono audio by default. Use the Output Stereo toggle to convert to stereo for ComfyUI compatibility.

Node Parameters

  • Audio: Input audio (required)
    • Works with any audio format supported by ComfyUI's Load Audio node
    • Automatically accepts any sample rate
    • Converts stereo to mono automatically
  • Model: Model file to use (optional, defaults to available model)
    • Supports .bin and .safetensors formats
  • Output Stereo: Convert mono output to stereo (optional, default: False)
    • Duplicates mono channel for ComfyUI pipelines that require stereo input
  • Unload Model: Remove model from VRAM after processing (optional, default: False)
    • Frees GPU memory but slower on next run
  • Show Spectrogram: Generate before/after spectrogram comparison (optional, default: True)
    • Visualizes frequency improvements

Audio Format Support

NovaSR works with any audio format that ComfyUI supports:

Supported Formats:

  • WAV (.wav)
  • MP3 (.mp3)
  • FLAC (.flac)
  • OGG Vorbis (.ogg)
  • AIFF (.aiff)
  • And more via ComfyUI's Load Audio node

Audio Processing:

  • Input Sample Rate: Any (automatically resampled to 16kHz)
  • Input Channels: Mono or Stereo (stereo automatically mixed to mono)
  • Output Sample Rate: 48kHz (fixed)
  • Output Channels: Mono (default) or Stereo (when toggle enabled)

Tip: Enable Output Stereo toggle if your ComfyUI pipeline requires stereo audio. This duplicates the mono channel to both left and right channels.

How It Works

NovaSR is a revolutionary tiny model (only 50KB!) that can upscale muffled 16kHz audio into clear and crisp 48kHz audio. It uses:

  • Less than 10 tiny Conv1D layers
  • Snake activations based on BigVGAN
  • Extreme efficiency optimizations

Comparison with Other Models

| Model | Speed (Real-Time) | Model Size | |-------|------------------|------------| | NovaSR | 3600x | 48-53 KB | | FlowHigh | 20x | ~450 MB | | FlashSR | 14x | ~1000 MB | | AudioSR | 0.6x | ~2000 MB |

Use Cases

  • Enhancing TTS model quality with nearly 0 computational cost
  • Real-time enhancement of low-quality calls and audio
  • Restoring audio datasets
  • Improving voice quality in recordings

Credits

  • ComfyUI-NovaSR Plugin: https://github.com/Saganaki22/ComfyUI-NovaSR
  • Model on Hugging Face: https://huggingface.co/drbaph/NovaSR
  • Original NovaSR by Yatharth Sharma: https://github.com/ysharma3501/NovaSR

License

This project follows the license of the original NovaSR repository.