ComfyUI LoRA adaLN Patcher Node

ComfyUI LoRA adaLN Patcher Node

A simple but powerful custom node for ComfyUI that patches LoRA models by adding dummy adaLN_modulation_1 weights. This solves compatibility errors when using LoRAs with newer model architectures that expect these keys to be present in the final_layer.

🚀 The Problem It Solves

Some modern U-Net architectures and updated ComfyUI workflows require weights for an Adaptive Layer Normalization (adaLN) module in the final block of the U-Net. However, some LoRAs were trained on architectures that did not have this module.

When you try to load such a LoRA, you might encounter a KeyError because the loader cannot find expected keys like:

• lora_unet_final_layer_adaLN_modulation_1.lora_down.weight
• lora_unet_final_layer_adaLN_modulation_1.lora_up.weight

This node fixes the issue by adding zero-filled "dummy" weights for these missing keys, making the LoRA compatible without altering its artistic output.

✨ Features

• Automated Patching: Detects and patches LoRAs that have final_layer.linear weights but are missing the corresponding adaLN weights.
• Safe & Non-Destructive: Creates a new, patched file (e.g., my_lora_patched.safetensors) and leaves the original file untouched.
• Smart Detection: Intelligently skips files that are already patched or do not require patching.
• Easy Integration: Works as a standard ComfyUI node, fitting seamlessly into your workflow.

📦 Installation

1. Navigate to your ComfyUI custom_nodes directory:

   cd ComfyUI/custom_nodes/
   

2. Clone this repository:

   git clone https://github.com/Franklyc/comfyui-lora-adain-patcher-node.git
   

3. Restart ComfyUI. The node will be available under the loaders/patchers category.

🛠️ How to Use

1. In ComfyUI, add the node by right-clicking and selecting: Add Node > loaders/patchers > Patch LoRA (adaLN).
2. From the lora_name dropdown, select the LoRA file you want to patch.
3. Click Queue Prompt to run the patching process.

Outputs

• patched_lora_name: The filename of the newly created (or existing) patched LoRA.
• patch_status: A message indicating the result (e.g., "Success!", "Skipped", "No patch needed"). You can connect this to a Show Text node to see the status.

Important: After a LoRA is successfully patched, you must Refresh your ComfyUI browser page (or click the Refresh button in the UI) for the new _patched.safetensors file to appear in other LoRA loader dropdowns.

📜 Technical Details

The node loads the state dictionary of the selected LoRA and checks for the existence of weights under common prefixes for the final layer (lora_unet_final_layer, final_layer, etc.). If it finds linear projection weights (...linear.lora_...) but not the corresponding adaptive layer norm weights (...adaLN_modulation_1...), it creates zero-tensors with the correct shapes and adds them to the state dictionary before saving it as a new file.

Because the added weights are all zeros, they have no mathematical effect on the LoRA's output, serving only to satisfy the model loader's key requirements.

🙏 Acknowledgements

This tool is an implementation of a hotfix script, adapted for easy use within the ComfyUI ecosystem. based on: https://github.com/tazztone/patch_comfyui_nunchaku_lora

---

Feel free to open an issue or submit a pull request if you have suggestions for improvement!