LORA trained on 158 images of artist はる?, (haru_(konomi150) tag on gelbooru) - https://www.pixiv.net/en/users/4790435 Doesn't require trigger words, just load LORA in and it will work. Weight 1 is ok. Training and dataset parameters: Images are scrapped from gelbooru with their tags, haru_(konomi150) tag was then removed from all of them; Trained using kohya-ss LORA training GUI, rank 64, resolution 512x768, no regularisation folder, other settings were default; Trained on anything-v4.5-pruned-fp16 model; 28 epochs, 1 step per image.
|