So, I wanted to do pic picture with Katsuragi p- ehm... squirting on the viewer but couldn't get droplets on the "lens".
Then I wanted to do a Cyberpunk cityscale with rain drops on the viewer and still couldn't find a way...
So I say f- it, I'm gonna try to do a lora. I never did something like that. It's not super reliable, from my test you go to put it at 1.2-1.4 for it to work. Your characters might get wet in the process. negging wet clothes or web might help avoiding that!
V2
V2 works better with animu and comics I feel
It doesn't turn everybody into Katsuragi (but because I'm an idiot I only did katsu pictures to illustrate the lora 🤪)
I still recommend using it only during img2img upscaling but it has less stylistic impact on the picture.
V1
This lora was trained with a teeny-weeny little microset of pictures. I might redo one with more pics.
The lora doesn't work well with every checkpoint, it seems to work pretty nice with new mecha
I recommend only using the lora during the img2img upscaling, not during txt2img generation
Because of the tiny dataset, it seems the lora also turn girls into Katsuragi... oh well.