LoRA (Low-Rank Adaptation) is an efficient technique for fine-tuning large generative models like Stable Diffusion. It introduces trainable low-rank matrices to the pre-trained model, adjusting only a portion of parameters rather than retraining the entire model, thus achieving optimization for specific tasks at a lower computational cost. Compared to base models like SD1.5, LoRA models are smaller and easier to train.

The image above compares generation with the same parameters using dreamshaper_8 directly versus using the blindbox_V1Mix LoRA model. As you can see, by using a LoRA model, we can generate images in different styles without adjusting the base model.

We will demonstrate how to use a LoRA model. All LoRA variants: Lycoris, loha, lokr, locon, etc… are used in the same way.

In this example, we will learn how to load and use a LoRA model in ComfyUI, covering the following topics:

  1. Installing a LoRA model
  2. Generating images using a LoRA model
  3. A simple introduction to the Load LoRA node

​
Required Model Installation

Download the dreamshaper_8.safetensors file and put it in your ComfyUI/models/checkpoints folder.

Download the blindbox_V1Mix.safetensors file and put it in your ComfyUI/models/loras folder.

​
LoRA Workflow File

Download the image below and drag it into ComfyUI to load the workflow.

Images containing workflow JSON in their metadata can be directly dragged into ComfyUI or loaded using the menu Workflows -> Open (ctrl+o).

​
Complete the Workflow Step by Step

Follow the steps in the diagram below to ensure the workflow runs correctly.

  1. Ensure Load Checkpoint loads dreamshaper_8.safetensors
  2. Ensure Load LoRA loads blindbox_V1Mix.safetensors
  3. Click the Queue button, or use the shortcut Ctrl(cmd) + Enter to generate the image

​
Load LoRA Node Introduction

Models in the ComfyUI\models\loras folder will be detected by ComfyUI and can be loaded using this node.

​
Input Types

Parameter NameFunction
modelConnect to the base model
clipConnect to the CLIP model
lora_nameSelect the LoRA model to load and use
strength_modelAffects how strongly the LoRA influences the model weights; higher values make the LoRA style stronger
strength_clipAffects how strongly the LoRA influences the CLIP text embeddings

​
Output Types

Parameter NameFunction
modelOutputs the model with LoRA adjustments applied
clipOutputs the CLIP model with LoRA adjustments applied

This node supports chain connections, allowing multiple Load LoRA nodes to be linked in series to apply multiple LoRA models. For more details, please refer to ComfyUI Multiple LoRAs Example

​
Try It Yourself

  1. Try modifying the prompt or adjusting different parameters of the Load LoRA node, such as strength_model, to observe changes in the generated images and become familiar with the Load LoRA node.
  2. Visit CivitAI to download other kinds of LoRA models and try using them.