Name your model

Choose a memorable name for your model. Renaming is not possible after training.

Choose a unique trigger word

Ensure the trigger word is unique and not a real word.

Choose good photos

Upload 4-10 high-quality samples with a square aspect ratio and various angles.

Train your model

Training takes 15-30 minutes. You can leave the page and return later.

Generate images

Include the trigger word in prompts to activate your concept.

Advanced Settings

• Model Quality: Good (250 steps), Better (500 steps), Best (1000 steps)
• Subject Caption: Optional text for generated captions, e.g., ‘a photo of JohnDoe’
• LoRA Rank: Higher ranks capture more complex features; default is 16
• Automatic Subject Caption: Uses Llava v1.5 13B to automatically caption images