I want to warn you right away: I have little experience and this LoRa is far from perfect. It's unstable and unpredictable, like most weapon LoRa i've tried before, to be honest. Getting a decent result is either a gamble or manual correction. And you'll probably have to burn thousands of points to get it.
The most problematic part is the ammo box and the ammo belt. The bipod is also problematic; you can try add "Bipod" to the negative prompt.
The results are noticeably better on some models than on others, but it requires experiments.
Recommended weight: 0.8-1.0
There are two versions; V0.5 has a smaller dataset, but the results are sometimes better.
I'll describe my workflow in case it helps someone. First, I gamble with the low-priced set. Next, I use the acceptable results as a base image (weight 0.55-0.65) or use them in ControlNet (depth) (weight 0.7-1.0) and continue gambling with more steps and facefix. If some details don't turn out right, I add or erase them in the base image using the "Paint" feature. Even a simple green box or black stripes drawn with a mouse or finger can fix problematic details.
