LoRA Training & Usage

Train custom LoRA models and use them with Get3W's image and video generation models.

What is LoRA?

LoRA (Low-Rank Adaptation) is a technique for fine-tuning AI models with your own images. This allows you to:

• Generate images in a specific style
• Create consistent characters or objects
• Customize model behavior for your use case

Training a LoRA

Prerequisites

• 10-20 high-quality training images
• A Get3W account with credits
• Images should be consistent in subject matter

Via Web Interface

1. Go to get3w.com/models
2. Search for "LoRA Trainer"
3. Upload your training images
4. Configure training parameters
5. Start training

Via API

curl -X POST "https://api.get3w.com/api/v3/get3w/flux-dev-lora-trainer" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "images": [
      "https://example.com/image1.jpg",
      "https://example.com/image2.jpg"
    ],
    "trigger_word": "my_style",
    "steps": 1000,
    "learning_rate": 1e-4
  }'

Training Parameters

Parameter	Description	Default
images	Array of training image URLs	Required
trigger_word	Word to activate the LoRA style	Required
steps	Number of training steps	1000
learning_rate	Training learning rate	1e-4
resolution	Training resolution	512

Using a Trained LoRA

Once training completes, you receive a LoRA model URL. Use it with compatible models:

curl -X POST "https://api.get3w.com/api/v3/get3w/flux-dev-lora" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A portrait in my_style, detailed, high quality",
    "lora_url": "https://cdn.get3w.com/loras/your-lora.safetensors",
    "lora_strength": 0.8
  }'

LoRA Parameters

Parameter	Description	Default
lora_url	URL to your trained LoRA	Required
lora_strength	How strongly to apply LoRA	0.8
prompt	Must include trigger word	Required

Tips

• More images = better results — Use 15-20 diverse images of your subject
• Consistent quality — Use high-resolution, well-lit images
• Trigger words — Use unique trigger words to avoid conflicts
• Strength tuning — Start with 0.8 and adjust up or down

Compatible Models

LoRA training and usage is available for:

• FLUX Dev LoRA
• FLUX Schnell LoRA
• Wan 2.1 / 2.2 LoRA variants
• Stable Diffusion LoRA variants

Next Steps

• How to Upload Files — Upload training images
• How to Generate an Image — Use your LoRA