Skip to content

Lora Loader

Documentation

  • Class name: AV_LoraLoader
  • Category: Art Venture/Loaders
  • Output node: False

The AV_LoraLoader node is designed to load and apply LoRA (Low-Rank Adaptation) models to given models and CLIP instances, optionally overriding the default LoRA model with a specified one and enabling or disabling the loading process. This functionality enhances model customization and flexibility in processing, allowing for dynamic adjustments and optimizations based on specific requirements or preferences.

Input types

Required

  • model
    • The model parameter represents the neural network model to which the LoRA adjustments will be applied, serving as the base for modifications.
    • Comfy dtype: MODEL
    • Python dtype: torch.nn.Module
  • clip
    • The clip parameter signifies the CLIP model that will be adjusted alongside the main model, allowing for synchronized enhancements in processing capabilities.
    • Comfy dtype: CLIP
    • Python dtype: torch.nn.Module
  • lora_name
    • Specifies the name of the LoRA model to be loaded and applied, enabling targeted modifications to the base model and CLIP instance.
    • Comfy dtype: COMBO[STRING]
    • Python dtype: str
  • strength_model
    • Specifies the strength of the LoRA adjustment to be applied to the model, allowing for fine-tuned control over the adaptation process.
    • Comfy dtype: FLOAT
    • Python dtype: float
  • strength_clip
    • Specifies the strength of the LoRA adjustment to be applied to the CLIP model, enabling precise customization of the enhancement.
    • Comfy dtype: FLOAT
    • Python dtype: float

Optional

  • lora_override
    • Allows for the specification of an alternative LoRA model to override the default, providing flexibility in model customization.
    • Comfy dtype: STRING
    • Python dtype: str
  • enabled
    • A boolean flag that determines whether the LoRA loading and application process should be executed, offering control over the modification workflow.
    • Comfy dtype: BOOLEAN
    • Python dtype: bool

Output types

  • model
    • Comfy dtype: MODEL
    • Returns the modified neural network model with applied LoRA adjustments, reflecting the enhancements or customizations made.
    • Python dtype: torch.nn.Module
  • clip
    • Comfy dtype: CLIP
    • Returns the modified CLIP model with applied LoRA adjustments, showcasing the enhancements or customizations made to processing capabilities.
    • Python dtype: torch.nn.Module

Usage tips

Source code

class AVLoraLoader(LoraLoader):
    @classmethod
    def INPUT_TYPES(s):
        inputs = LoraLoader.INPUT_TYPES()
        inputs["optional"] = {
            "lora_override": ("STRING", {"default": "None"}),
            "enabled": ("BOOLEAN", {"default": True}),
        }
        return inputs

    CATEGORY = "Art Venture/Loaders"

    def load_lora(self, model, clip, lora_name, *args, lora_override="None", enabled=True, **kwargs):
        if not enabled:
            return (model, clip)

        if lora_override != "None":
            if lora_override not in folder_paths.get_filename_list("loras"):
                print(f"Warning: Not found Lora model {lora_override}. Use {lora_name} instead.")
            else:
                lora_name = lora_override

        return super().load_lora(model, clip, lora_name, *args, **kwargs)