Trained on dataset of 94 (I excluded from the dataset some images that had noise and imperfections, and I tried to balance it by increasing the color images) topaz photo AI enhanced images x 15 repeats and BLIP manually edited captions and with 1410 regularisation images
Trained on dataset of 94 (I excluded from the dataset some images that had noise and imperfections, and I tried to balance it by increasing the color images) topaz photo AI enhanced images x 15 repeats and BLIP manually edited captions and with 1410 regularisation images
Kohya settings:
Optimizer: Dadaptation
LR Scheduler: Costant
LR warmup: disabled
Learning rate: 0.5
Text encoder learning rate: 0.5
Unet learning rate: 0.5
Network Rank (Dimension): 128
Network Alpha: 1
Clip skip: 1
Noise offset: 0.01
Epoch: 8
Steps: 22560