2-7 LoRA Training

Train your own LoRA models for AI art generation. Learn about dataset creation, image preprocessing, tagging, and publishing your trained LoRA. Let's dive in!

Page entry: Click on "Train" in the upper left corner of the homepage to access the style training page.

Operation Process

1. Create Dataset: Upload dataset images, ensuring diversity in angles, lighting, backgrounds, etc., with a recommended count of around 30 images.

2. Image Preprocessing: Crop images, add taggers, and include trigger words.

   • Cropping Mod: Recommend focus cropping.

   • Size: Choose according to output requirements; Square: 512 * 512; Portrait: 512 * 768; Landscape: 768*512.

   • Tagging Algorithm: Recommend Deepbooru.

   • Tagging Threshold: Recommend 0.6.

   • Trigger Words (Optional): A word that invokes Lora.

3. Choose preset parameters based on the trained Lora.

4. Enter model preview prompt words, mainly to review Lora's outputs during training.

5. Input the dataset name and click "Train Now".

*For detailed instructions on Tagging and advanced training parameter settings, click here.

3-2 LoRA Training (Advance)

Publish

1. Click on Train -> LoRA -> Publish to view the trained Lora.

2. Click on Publish, edit the Lora information, and fill in the relevant information as prompted, including LoraName, Tags (optional), Lora Cover, Model Usage, and Model Permissions.

Paid: When others use the model you published, you can receive corresponding Credits and get 70% of the Amount set.

3. Edit Version:

Includes: VersionName, Base Model, Trigger Words, Version Introduction.

4. Add image.

Upload the cover image for the LoRA and other showcase images.

5. Return to the LoRA page and click "Public."

Edit the Lora again

Click on the Lora details page, then the three dots in the bottom right corner, and select "Edit.”

Take the Lora offline

After publishing the Lora, click on the Lora details page, then on the three dots in the bottom right corner, and select "Private.”

Training the same Lora

1. Click on "Lora Training" on the homepage.

2. Select any Lora template, then click on "Train" in the top right corner.

*You can modify parameters, datasets, etc., based on the same template.