Testing style training with Fizgig ...it's awesomeSome image quality tags will render this lora useless . Try to increase strength to max 1.75 .I didn't bother with captions just ran with what qwen 3 vl 4b gave , ran 30 epochs...i'll train further another 70 to see what is what. It's not consistent , needs more testing or more images or something...this is my first time training a style so ...yeahthe training data is also available for dl if you are curious

BastardStyle

steps: 100%|██████████| 960/960 [1:46:37&lt;00:00, 6.66s/it, avr_loss=0.1155]34 images from Bastard v2 ill

Files with SHA256: 09adda3732b2f6e631f11f0893468de3832adce34081e6c201623ed5ef72360d