Datasets

Training a LoRA takes three steps, found in the “Training” space at the top:

Step	Description
Datasets	Manage the folders of training images
Tagging	Write tags and descriptions for each image; see Tagging
Train	Choose a base model and settings and start training; see Training LoRA

The dataset list

“Datasets” shows each dataset as a card with:

• The number of images, and how many are tagged
• The trigger word and repeats per image
• The most recent training
• Notes

Click a card to open it in “Tagging”.

Adding and removing

• Press “+ Add dataset” and choose the folder with your training images (you can also create a new folder in the folder window).
• Pressing “×” on a card twice removes it from the list. This only removes it from the list; the files in the folder are not deleted.

Adding images from the library

Select the images to train on in the library and press “Add to dataset” in the action bar:

1. Choose a dataset from the list, or enter a folder path directly (it’s created if it doesn’t exist), or press “Browse…”.
2. Press “Copy N”.

The images are copied into the dataset folder; the originals don’t move. Afterwards, “Go to tagging” takes you straight to tagging.

Dataset settings

In the left panel of “Tagging”:

Item	Description
Repeats per image	How many times each image is repeated in training. Shared with the training page; changing it on either side updates both
Notes	Purpose, source, versions trained and so on
Trigger word	Added to the front of every image’s tags; see Tagging

While reviewing, “Set as cover” changes the dataset card’s cover (the first image by default).

Training history

The bottom of the “Tagging” left panel lists this dataset’s training history: name, file name, base model, date and epochs; trainings stopped early are recorded too. “Use these settings” brings that run’s full settings to the training page to train again.