ComfyUI-Musubi-Tuner-LoRA-Loader

ComfyUI Musubi-Tuner LoRA Loader

A ComfyUI custom node that enables direct loading of musubi-tuner based LoRAs without manual format conversion. The conversion happens automatically in-memory during node execution.

Features

• On-the-fly Conversion: Automatically converts musubi-tuner LoRA format to ComfyUI-compatible diffusers format
• No File Creation: All conversion happens in-memory, no temporary files needed
• Multi-Format Support:
  • Qwen-Image LoRAs
  • Z-Image Turbo LoRAs
  • Hunyuan Video 1.5 LoRAs
  • FLUX LoRAs
  • Wan2.1 LoRAs
• ComfyUI v3 Schema: Uses the latest ComfyUI extension API

Installation

1. Navigate to your ComfyUI custom nodes directory:

   cd ComfyUI/custom_nodes/
   

2. Clone this repository:

   git clone https://github.com/vjumpkung/ComfyUI-Musubi-Tuner-LoRA-Loader.git
   

3. Restart ComfyUI or reload custom nodes

Usage

1. Find the node in ComfyUI under the name "musubi-tuner LoRA Loader"
2. Connect a MODEL input
3. Select your musubi-tuner format LoRA from the dropdown
4. Adjust the strength_model parameter (default: 1.0, range: -100.0 to 100.0)
5. Connect the MODEL output to your workflow

How It Works

The node converts musubi-tuner LoRA format to diffusers-compatible format using a comprehensive mapping system:

1. Format Detection and Module Name Mapping

The conversion automatically detects the LoRA format based on key patterns:

• Qwen-Image: Uses predefined key mappings for transformer blocks
• Z-Image: Detects .attention.to. and .feed.forward. patterns
• Hunyuan Video/FLUX: Detects double.blocks. and single.blocks. patterns
• Wan2.1: Detects .cross.attn. and .self.attn. patterns

2. Key Transformation

Converts musubi-tuner naming conventions to diffusers format:

• Replaces underscores with dots for module hierarchy
• Applies format-specific fixes (e.g., to.q → to_q, img. → img_)
• Maps to diffusers structure: lora_down → lora_A.weight, lora_up → lora_B.weight

3. Alpha Scaling

Applies LoRA alpha scaling to weights:

• Scales both down and up weights by sqrt(alpha / dim)
• Ensures proper weight magnitude for model integration

Technical Details

• API Version: Built with ComfyUI v3 Schema API (comfy_api.latest)
• Base Class: Implements io.ComfyNode with define_schema() and execute() methods
• Extension System: Uses ComfyExtension and comfy_entrypoint() for registration
• Optimization: Returns original model immediately if strength_model == 0
• Dependencies: torch, tqdm, comfy.sd, comfy.utils, folder_paths

Credits

Conversion logic adapted from musubi-tuner by kohya-ss

License

This project is provided as-is for use with ComfyUI.