AudioDriven Latent Space Tools for ComfyUI

Generate dynamic latent noise patterns driven by audio analysis.

Nodes

Librosa Audio Analysis

Analyzes audio files using various methods:

• Onset detection (note starts)
• Spectral content
• Beat tracking
• Tempo analysis
• MEL spectrograms

Audio To Noise Parameters

Converts audio analysis into noise parameters:

• Intensity based on audio energy
• Grain size from spectral features
• Persistence from temporal changes

Audio Noise to Latent

Converts noise parameters to latent space noise using:

• Gaussian noise
• Salt & pepper noise
• Perlin noise

Advanced Audio Noise Patterns

Creates musical visualization-style patterns:

• Simplex noise
• Cellular patterns
• Fractal Brownian Motion (FBM)
• Wave patterns
• Domain warping

Usage

1. Input audio file through Librosa Analysis
2. Choose analysis type (onset/tempo/mel/etc.)
3. Convert to noise parameters
4. Generate latent noise
5. Connect to ksampler
6. Use with low denoising (<0.5) for audio-visual blending

WARNING: Advanced Audio Noise Patterns Node

Typical image counts: Average measurements by analysis type (from 3s, 11.65s, and 89s samples)

Default: huge (~122K for 89s)

Mel/spectral/tempo: high (~7.6K for 89s)

Onset: moderate (60-500 depending on complexity) 

Beat: low (14-168 based on tempo) 

Second: consistent (~length in seconds) 

Half_second: consistent (~2x seconds) 

Installation

Download or clone this repository to your ComfyUI\custom_nodes folder

This node is part of a ongoing solo project to integrate music analysis. plz enjoy.