V2

I've discarded the entire original dataset and collected many different images.

It's superior to the previous version and performs well in full-body mode, from what I've tested.

It operates with a force between 0.6 and 1

Perhaps the only flaw is the gag; if it's a Bit Gag, it makes the face look like it's biting, although it's random.

It also almost always maintains the same arm position, but by forcing it a little, it can be changed

This LoRa does not support "bridle, leash, Reins, Horseback riding, Stirrups" They won't be generated well

V1

Also put "bridle, leash, Reins, Horseback riding, Stirrups" in negatives, since it generates all that and can generate half-done things.

In full body images it tends to break the face (or I'm bad at doing full body images), but in "cowboy shot" images upwards, everything is fine.

Additionally, it tends to generate the occasional metal/wooden structure if it is full-body.