This LoRA enhances the concept "holding gun" and derivates.
Enchances and not makes perfect, far from it!
The holding concepts are messed beyond hope and would require a new whole base model.
Some models aren't trained at all with ranged weapons and will fail consistently.
Prompting: holding ranged weapon, detailed realistic weapon type, action, usual prompt
Example: holding ranged weapon, beretta 92, aiming at viewer
While it should enhance various concepts, the LoRA is especially training on the following concepts:
Weapon type:
beretta 92
hk416
sniper riffle
assault riffle
revolver
submachine gun
sig sauer
kalashnikov riffle
kechler & koch
m1911
handgun (global type)
riffle (global type)
Extras for weapons:
suppressor
scope
Extra actions:
trigger discipline
finger on trigger
aiming at viewer
Beretta 92 is the most trained gun model and should give the best results.
You can use other characters than Eula (model isn't trained at all on her) but well, you won't have Eula in your picture and that's pitiful.
Technical aspects:
This LoRA is just a carefully, manually, painfully tagged/pruned dataset. of 183 images, both anime and realistic.
It can be improved and maybe will be. Keep in mind a simple LoRA can't hold on his shoulders all the messed up concepts of a model.
At some points, a whole base model needs to be trained. Let's hope for SD3!