Create New LoRA
This will be the name of your trained model that you can select when generating images
Name for your LoRA model (max 50 characters)
Internal name for your training dataset. You can use this to organize your training data.
Name for your training dataset (max 50 characters)
This unique word activates your style in prompts.
This word will trigger your style in prompts. Use a unique word that does not conflict with other concepts. (max 20 characters)
Higher steps produce better quality but cost more tokens and take longer to train. 2000-3000 steps is a good balance for most styles.
Recommended: 2000-4000 steps. Minimum 500, maximum 4000. More steps may improve quality but take longer and cost more tokens.
These prompts are used to generate sample images during training to show progress. Always include your trigger word in each prompt.
Add prompts that describe your style. Include your trigger word in each prompt. Maximum 10 prompts, 500 characters each.
Upload 10-20 high-quality images that represent your style. Supported formats: JPG, PNG, WEBP. For best results, use consistent images with similar style elements.
Before You Submit
- Training a LoRA model typically takes 1-3 hours depending on steps and image count
- You will be able to monitor training progress on the training jobs page
- Once training is complete, your model will be available in the image generation page
- Higher step counts produce better results but consume more tokens