LoRA Tutorial

Use LoRA models for style and content control.

What is LoRA

LoRA (Low-Rank Adaptation) is a lightweight model that modifies the base model's behavior without changing the entire model.

Use Cases

• Apply specific art styles
• Character consistency
• Special effects
• Concept modification

LoRA Files

Format: .safetensors or .pt

Location: ComfyUI/models/loras/

Basic Usage

Step 1: Load LoRA

Node: LoraLoader
- Select your LoRA file
- Set strength (typically 0.5-1.0)

Step 2: Connect

[CheckpointLoader] → [LoraLoader] → [CLIP/Model outputs]

Step 3: Use in Prompt

Some LoRAs require trigger words:

# Check LoRA description for trigger
trigger-word, another-trigger

# Or use embedding style
embedding:my-lora-name

LoRA Parameters

Strength

Value	Effect
0.1-0.3	Subtle influence
0.4-0.7	Balanced
0.8-1.2	Strong effect
>1.5	May cause artifacts

Multiple LoRAs

# Use multiple LoRAs
<lora:style1:0.8> <lora:style2:0.5>

Popular LoRA Types

Style LoRA

LoRA	Style
anime	Anime style
portrait	Photorealistic portrait
watercolor	Watercolor painting
pixel	Pixel art

Character LoRA

• Maintain consistent character
• Use trigger words from model page
• Adjust strength for best results

Concept LoRA

• Add specific concepts
• Custom poses
• Special effects (lighting, atmosphere)

SDXL LoRA

SDXL models require SDXL-specific LoRAs:

ComfyUI/models/loras/
  sdxl-compatible-lora.safetensors

SDXL LoRA Settings

Parameter	Value
Model Strength	0.5-1.0
CLIP Strength	0.5-1.0
Both	Match for best results

Advanced: Custom LoRA Training

Training Requirements

• 10-50 sample images
• Consistent subject/style
• Quality over quantity
• GPU with 8GB+ VRAM

Training Process

1. Prepare dataset
2. Configure training parameters
3. Train model
4. Test and iterate

Tips

1. Start low: Try 0.5-0.7 strength first
2. Check trigger words: Read LoRA description
3. Combine with prompts: Describe desired result
4. Test variations: Adjust strength and prompts

Troubleshooting

Issue	Solution
No effect	Check if LoRA loaded correctly
Too strong	Lower strength value
Artifacts	Reduce strength or try different LoRA
Not matching	Use trigger words

Next Steps

• ControlNet - Structure control
• Inpainting - Fix specific areas
• SDXL Workflow - SDXL features