ComfyUI LoRA Usage Tutorial: A Step-by-Step Guide

LoRA (Low-Rank Adaptation) is a lightweight fine-tuning technique widely used in stable diffusion models to add specific styles, characters or objects. Compared to full model fine-tuning, LoRA files are small (usually a few MB to tens of MB), load quickly, and are highly flexible—making them perfect for customizing image generation in ComfyUI.

before using

you should know: lora have impact on image

step 1: load lora workflow

Enter the templates folder → search qwen

step 2: download model files

due to the fact that lora based on z-image-turbo is short ,so i recommend you to download dreamshaper model to use lora which has lots of trained loras. here is model linkblindbox_V1mix.safetensors,and put them in ComfyUI/models/checkpoints/ andComfyUI/models/Lora/ folder. or just download by click download button

Step 3: Use LoRA in ComfyUI Workflow

loadlora node

Step 4: Advanced Tips

4.1 Use Multiple LoRAs Together

You can stack multiple LoRAs in a single workflow by adding multiple LoraLoader nodes. Follow these rules:

• Connect the first LoraLoader to the base model, the second LoraLoader to the first one’s outputs, and so on (chain them).
• Reduce the strength of each LoRA (e.g., 0.4–0.6 per LoRA) to avoid conflicting effects and distortion.
• Prioritize LoRAs: Place more important LoRAs (e.g., character) earlier in the chain, and style LoRAs later. Example chain: CheckpointLoader → Character LoRA → Style LoRA → KSampler

4.2 LoRA Compatibility with different base models

• You should know that lora is trained on a specific base model.If the lora don't match the base model, the generated image will be bad.
• lora based on new advanced models like z-image-turbo is less than old models like sdxl dreamshaper.So you need to consider the base model when using lora.
• of course, you can train lora by yourself.

Write in the end: most of loras is used for add specific style of image. But some loras can help accelerating the generation process like just 4 steps to generate a good image. So if you meet the similar situation, calm dowm.

Common Issues & Troubleshooting

Issue 1: LoRA Doesn’t Appear in the LoraLoader Dropdown

Possible fixes:

• Ensure the LoRA file is in ComfyUI/models/Lora/ (not a subfolder).
• Restart ComfyUI—new LoRA files won’t be detected until ComfyUI is restarted.
• Check the file extension: ComfyUI prefers .safetensors, but .ckpt should also work. Avoid .pt files.(by the way, .checkpoint is old and safetensors is new and safer.They are both file types. )
• Verify the LoRA is compatible with your base model (e.g., SD 1.5 LoRA → SD 1.5 base model).

Issue 2: LoRA Effect Is Not Visible

Possible fixes:

• Add the required keywords to your prompt (check the LoRA’s download page).
• Increase strength_model and strength_clip (try 1.1–1.2).
• Ensure the LoraLoader is connected correctly (model → sampler, clip → text encode).
• Check if the LoRA is for the correct base model (e.g., SDXL LoRA won’t work with SD 1.5).

Issue 3: Image Distortion or Glitches

Possible fixes:

• Decrease strength_model and strength_clip (try 0.7–0.9).
• If using multiple LoRAs, reduce the strength of each (total strength should not exceed 1.5).
• Avoid conflicting LoRAs (e.g., two character LoRAs or two conflicting style LoRAs).
• Update your base model and ComfyUI to the latest version.

Issue 4: ComfyUI Crashes When Loading LoRA

Possible fixes:

• The LoRA file may be corrupted—re-download it from a trusted source.
• The LoRA file is too large (uncommon, but some LoRAs exceed 100MB)—try a smaller LoRA.
• Your GPU may not have enough VRAM—reduce image dimensions or close other applications.