I saw this fixhands lora getting shilled, seems meh like most loras of these types.
>>86199425For loras you want as many images as you can with as much variation of pose and style as possible as long as they're at a minimum level of quality and they accurately represent the character. A lot of people will caption a dataset themselves if given because either A: They don't clean up tags that much so it's a one-click process or B: They clean up tags so much that they probably have specific ways they like to do it and a different person might not do it correctly to their standards
