ComfyUI Node: Load UNET + LoRA (Model Only)

Load UNET + LoRA (Model Only):

The Sage_UNETLoRALoader node is designed to enhance the functionality of an already loaded UNET model by applying a LoRA (Low-Rank Adaptation) stack. This node is particularly useful for AI artists who want to modify or fine-tune their UNET models without altering the CLIP components. By integrating LoRA, the node allows for efficient model adaptation, enabling users to achieve desired artistic effects or model behaviors with minimal computational overhead. The primary goal of this node is to provide a streamlined process for applying model shifts and LoRA stacks, thereby enhancing the flexibility and adaptability of the UNET model in creative workflows.

Load UNET + LoRA (Model Only) Input Parameters:

unet_info

The unet_info parameter is essential as it provides the necessary information about the UNET model that you wish to modify. This input is crucial for the node to identify and access the correct model file. It typically includes details such as the model's file path and configuration settings. Ensuring that this information is accurate is vital for the successful execution of the node, as it directly impacts the model's loading and subsequent modifications.

lora_stack

The lora_stack parameter is an optional input that allows you to specify a stack of LoRA configurations to be applied to the UNET model. This stack can include multiple LoRA settings, each contributing to the overall adaptation of the model. By providing a lora_stack, you can fine-tune the model's behavior to better suit your artistic needs. If not provided, the node will proceed without applying any LoRA modifications, maintaining the model's original configuration.

model_shifts

The model_shifts parameter is another optional input that enables you to apply specific shifts or adjustments to the model. This can include changes in model parameters or weights that further refine the model's output. Like the lora_stack, this parameter is optional, and if not specified, the node will execute without applying any additional shifts, preserving the model's default state.

Load UNET + LoRA (Model Only) Output Parameters:

out_model

The out_model output represents the modified UNET model after the application of the LoRA stack and any specified model shifts. This output is crucial as it provides the final model that you can use for your creative projects, reflecting all the adaptations and enhancements made during the node's execution.

out_lora_stack

The out_lora_stack output provides the LoRA stack that was applied to the UNET model. This output is useful for verification purposes, allowing you to confirm the specific LoRA configurations that were used in the model adaptation process.

keywords

The keywords output is a string that contains relevant keywords associated with the applied LoRA stack. These keywords can be helpful for documentation or further processing, as they encapsulate the essence of the modifications made to the model.

Load UNET + LoRA (Model Only) Usage Tips:

• Ensure that the unet_info parameter is correctly configured with accurate model path details to avoid loading errors.
• Experiment with different lora_stack configurations to achieve various artistic effects and model behaviors.
• Utilize the model_shifts parameter to make precise adjustments to the model, enhancing its adaptability to specific tasks.

Load UNET + LoRA (Model Only) Common Errors and Solutions:

UNET info is missing or invalid.

• Explanation: This error occurs when the unet_info parameter is not provided or contains incorrect information, preventing the node from loading the UNET model.
• Solution: Verify that the unet_info parameter is correctly set with the appropriate model path and configuration details.

Timestamp update failed for loras

• Explanation: This warning indicates that the node encountered an issue while updating the timestamp for the LoRA files, which may affect version tracking.
• Solution: Check the file paths and permissions for the LoRA files to ensure they are accessible and writable by the node.