/vt/ - Virtual Youtubers

>>76580154
I think the general explanation is that the dataset used to train AI models (which are usually pulled straight from boorus) never have tags that describe in detail how hands or fingers are featured in the image
as a result the AI model just has a general idea of how to recreate hands and doesn't really "understand" or parse that fingers are supposed to look one way from behind and another way when seen from the side and another way when in a grip and so on