>>2750

That's a good question to which I don't have a definitive answer. But from what I understand there are multiple problems that will make it hard to train a working Lora for generic stabbing. You need to either focus on a single type of weapon or a specific variety of weapons (like I'm trying to do for just pussy stabbing), otherwise at least the hilt/handle if not the whole weapon will in most cases be a melty unrecogniseable blob of AI weirdness. And therefore for each kind of weapon you will need a certain amount of pics in the training data. (at least 20) Then you'd need a comparable amount of pics with all kinds of camera-angles/poses (unless you want one of those Loras that can only do one angle/pose), a bunch of pics to cover for stabbing or slashing different bodyparts and for whatever else interaction between the characters you want (for example holding the stabbed subjects hair from behind vs. holding the stabbed subjects arms from behind). I think there would be even more things you'd need to consider and therefore cover in the dataset, but all this is just laymans speculation on my part.