🌊hua_gradio_Lora Loader

Hua_LoraLoader (🌊hua_gradio_Lora Loader) is the full-model-and-CLIP LoRA loader in the ComfyUI_to_webui pack. It's the classic LoraLoader under a hua name - take a MODEL and CLIP, apply a LoRA, hand back both modified - with a name label bolted on. It's the one to use when you want a LoRA applied inside your packaged graph but you don't need the user to choose which LoRA in the webui. If you do want that dropdown, that's the ModelOnly variant's job.

The mechanism is stock ComfyUI LoRA loading with a small cache: it looks up the file by lora_name, loads it once with load_torch_file, and remembers it so consecutive runs with the same file skip the reload. Strengths work the standard way - strength_model and strength_clip both default to 1.0 and can go negative (useful for "un-LoRA-ing" or style subtraction), and if both are zero it returns the model untouched, which is a nice free bypass.

Inputs, all required:

• model - MODEL from your checkpoint/UNet loader.
• clip - CLIP from the same loader.
• lora_name - dropdown from models/loras.
• strength_model - how hard the LoRA pushes the diffusion model; default 1.0, range −100 to 100.
• strength_clip - same, for the text encoder side; default 1.0.
• name - label.

Outputs: MODEL and CLIP, which you chain forward (LoRA nodes can stack: output of one into the next).

Install: ComfyUI Manager → "ComfyUI_to_webui", or

cd ComfyUI/custom_nodes
git clone https://github.com/kungful/ComfyUI_to_webui.git

then restart. No model files come with the pack; it lists whatever LoRAs you've already placed in models/loras.

The thing to understand is the division of labor in this pack. The README explicitly marks Hua_LoraLoader as "not added to components" - meaning the Gradio frontend does not render a selector for it. It's a graph-side node: you pick the LoRA once while building the workflow, and it stays fixed in the packaged UI. The 🌊hua_gradio_Lora Model Only node, by contrast, is the dynamic one that spawns a dropdown in the frontend. So: static style baked into the package → use this one (or just use the stock loader, honestly - the only difference is the label). User-facing model choice → the ModelOnly node. That split isn't obvious from the node names, so it's worth remembering.