>>2905741They are shitty datasets because remember it was for 1.5 lora, you could train that shit with 2 images
The reason I would have not retrained it is it would require to make good gems and use them for the new training.
I still have them and don't mind to share, Im just warning because you may be very dissapointed, then again because on 1.5 the quality standards were lower and the training required very few images
https://files.catbox.moe/72bcbj.ziphttps://files.catbox.moe/tawxl6.zip 