How kijai/comfyui-mmaudio works
Main Components
a. External Interface Layer:
• ComfyUI Integration Node(s)
• Model Loading Interface
• File System Interface (for model storage)
b. Core Processing Components:
• Audio Processing Pipeline
• Model Management System
• Extension Modules
c. External Dependencies:
• ComfyUI Framework
• Hugging Face Model Repository
• NVIDIA BigVGAN Models
Top Level (User Interface)
• It to "Node Management System" (nodes.py)
Middle Layer (Core Processing)
• Model Loading
• Model Storage
• Configuration Management (configs/)
• Autoencoder System (mmaudio/ext/autoencoder)
• BigVGAN Processing (mmaudio/ext/bigvgan_v2)
• Synchformer Processing (mmaudio/ext/synchformer)
Bottom Layer (Infrastructure)
• External Model Storage
• File System Interface
• Download Management System
Use these visual elements:
• Use hexagons for processing modules
• Blue: Core components
• Green: Processing modules
• Yellow: External interfaces
• Gray: Storage systems
Special Considerations:
Modular nature of the system by clearly separating
• The extension system (mmaudio/ext/)
• The core model components (mmaudio/model/)
• The utility functions (mmaudio/utils/)
Configuration flow
• From config files (configs/)
• Through the model initialization
• To the processing pipeline
Audio processing pipeline
• Input handling
• Model processing
• Output generation
A special section for the CUDA optimizations in bigvgan_v2
• Component types
• Abbreviations used