Models struggle to produce the minor complex variations involved in firearms manipulation. This lora does not fix that, but it does improve it somewhat for handguns. Sadly the results are not great and you will still need a mix of cherrypicking seeds, regional prompting and controlnets to get satisfactory results. Perhaps this will inspire someone else to produce a better lora.
tags of note:
cocking gun, locked slide, smoking barrel, reloading, magazine (weapon), holding gun,
known issues:
struggles to correctly connect arms to hand in newly learned gun cocking positions
strong concept bleed problems (eg. "smoking barrel" adding cigarette smoking and/or wooden barrel)
high lora strength required for adequate results has negative effects on overall image quality (ie. overbaked)
Because the model does enhance understanding of poses you can use an iterative process of controlnet editing to go from an initial generation with serious flaws to something a bit better.
