Skip to main content
This node creates a hook model as a LoRA (Low-Rank Adaptation) by loading checkpoint weights and applying strength adjustments to both the model and CLIP components. It allows you to apply LoRA-style modifications to existing models through a hook-based approach, enabling fine-tuning and adaptation without permanent model changes. The node can combine with previous hooks and caches loaded weights for efficiency.

​
Inputs

ParameterDescriptionData TypeRequiredRange
ckpt_nameThe checkpoint file to load weights from (select from available checkpoints)STRINGYesMultiple options available
strength_modelThe strength multiplier applied to the model weights (default: 1.0)FLOATYes-20.0 to 20.0
strength_clipThe strength multiplier applied to the CLIP weights (default: 1.0)FLOATYes-20.0 to 20.0
prev_hooksOptional previous hooks to combine with the newly created LoRA hooksHOOKSNo-
Parameter Constraints:
  • The ckpt_name parameter loads checkpoints from the available checkpoints folder
  • Both strength parameters accept values from -20.0 to 20.0 with 0.01 step increments
  • When prev_hooks is not provided, the node creates a new hook group
  • The node caches loaded weights to avoid reloading the same checkpoint multiple times

​
Outputs

Output NameDescriptionData Type
HOOKSThe created LoRA hooks, combined with any previous hooks if providedHOOKS
This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub
Source fingerprint (SHA-256): 8c0dd6b2e8e99e1d7dbc864aa802c0713842fb0d4ee018ea5cbedfb7896a770d