This is an experimental LoRA trained with the r34 tag "embedded_penetrable_sex_toy" in mind. It's pretty much a Proof Of Concept than a stable and user friendly LoRA to use, but I still hope you can have a good time trying it for yourself.
Even with a very small dataset, it do work with realistic and furry style.
For the Strength of this LoRA, it seem to work fine with 1, but I would recommend trying lowers values and work around trigger word to adjust the result
For all versions
The keyword is 'embedded_penetrable_sex_toy' but you can use specific keyword style like :
'human, sexdoll,' for a human character
'humanoid, robot girl, sexbot, mechanical, mechanical body,' for a robotic character, with a more raw mechanical interior
'robot furry, anthro, sexfurry,' to help furry and smoother skin generation
'robot furry, anthro, sexfnaf,' for a mechanical and rounder furry result, with more robotic attributes
For the color of the internal sextoy, you can use 'artificial vagina, <COLOR> fleshlight,' and custom it.
For Illustrious V2.0
There is different style of belly transformations, those are :
'inside view, mechanical body, womb window' : small womb window
'internal view, open body, exposed mechanical body, robotic body,' : larger window
'x-ray, inside view, see-through, see-through womb, transparent body, translucent body,' : largest internal exposition, you may want to use 'metallic skeleton' to increase the effect
In this last version the internal sextoy texture can be modified like so :
'solid fleshlight' : no texture visible, only rigs on surface
'transparent fleshlight' : visible inside texture
'see-through fleshlight' : fully clear internal texture
To increase the robotic aspect you can change weight of this new trigger word :
'grey metallic skeleton, shiny metallic skeleton' : chromed internal structure
'([light grey|grey|dark grey|black] metallic skeleton:0.7), translucent body part,' : all available nuance for the internal structure
Description
Changed training parameter for the same dataset as previous version
FAQ
Comments (9)
love this
could you add one that does realistic better?
Hey, sorry for the late response.
As for adding more realistic training dataset, I don't have that much idea.
I could try to generate with AI some example and refine it until I have something exploitable.
@Lazius that would be amazing man
@olivereaduzuh Sorry for the really long delay, I got a hard time until I found this Checkpoint https://civitai.com/models/572858/realistic-pony-photo-real I will upload examples. Those examples will be added to the dataset to train it and be less dependant of the checkpoint used
@Lazius thanks I'll give it a shot with this checkpoint but man am I getting spoiled by flux lately... can't wait to see pony + flux
@olivereaduzuh Yeah I just try Flux on my computer, it run but I didn't find how to train on 12GB, even with a civitai guide on this It still require 24GB vram :(
@Lazius I use a card that size too! I use gguc in comfy for output like everyone else, but for training I use onetrainer with the full size dev model and it works with the right settings
this took me weeks to figure out, here are my presets (point the trainer to the complete flux folder, not a safetensor):





