LoRA for a tank (or similar) that is part of a tentacle monster or other large living creature. Distinct from a stasis tank with tentacles because the tank itself is organic. Distinct from a tentacle pod because the tank is at least partially transparent.
IL_v1: Has as little style impact as I could manage while getting decent concept adherence. Less stable than I'd like, but more epochs didn't seem to help. Best at X+ content, to get an R rated image I had to pick from the best of 16 generations.
Main trigger: ttcltnk, tentacles, Inside container,
Optional (can help stabilize main trigger):
"submerged" - fills the tank with water
"partially submerged" - tank partway full with water or other fluid.
restrained - nothing unique, but can improve stability
against glass - nothing unique, but can improve stability
Known limitations: interacts oddly with "monster", very fond of the color pink, "submerged" tends to put the entire tank underwater, can struggle with long prompts. It's possible to make the LoRA work without "restrained", but it's a lot more reliable with it.
Description
IL_v1: Has as little style impact as I could manage while getting decent concept adherence. Less stable than I'd like, but more epochs didn't seem to help.
FAQ
Comments (5)
Excellent lora. I was struggling with concept adherence initially but adding "inside container, cramped, restrained" to the prompt did alot to resolve my issues. Really I think only restrained is doing anything there but I'll continue testing.
I forgot about that, I'll edit the model description.
"Inside container" is definitely the keyword for me. Tried it across multiple checkpoints too. I get very reliable images with it and more generic tentacle images without it. Against glass works which you demonstrated and you can also use "yellow-tinted glass". I've only found yellow works though so I presume this was picked up in a training image or two.
It can sometimes drift with too many prompts but simply increasing that tokens weight (plus ttcltnk) is often all I need to do.
@Kratnori That's odd, I trained on tags, not captions. Thanks for the help though.
Finally! Omg! Thx!



