About the LoRA
This is a model of the RED Mimi Sentry, who is a fan-made TF2 character.
It was trained to represent the actual 3D model, so usually you will get 3D-like results unless you tweak some parameters.
This model is:
trained on 3D images
trained on SFW (≈65%) and NSFW data (≈35%)
had some images upscaled for better training
(from V2) trained to work with other character colors (simpliest way is to change hair color to red, blue, blonde or green respectively, you can also try using red, blu, grn and yel tags)
This model is not:
trained on 2D images
trained on AI-generated images
Tips:
you may want to lower the weight to get more 2D-like images or to make it a bit more flexible in some cases (0.7-0.9 should do the trick while keeping the character recognizable)
if you want to get rid of some base elements, you may want to add the following tags to the negative prompt (or vice versa to make them appear):
arm cannon — changes the right hand to the arm cannon.
mechanical eye — changes the left eye to the mechanical one.
mini top hat — light bulb on the head.
white skin — makes it more white and android-like.
Note: since this is my first ever LoRA, it may have some issues, but the overall experience seems to be fine (especially from V2)
If you like this model
don't forget to give it a small like
share your generations with others and react to their generations
Description
Increased dataset size (now has about 4 times more images).
Removed irrelevant tags and tweaked existing ones (now it should be more flexible and precise).
Added blue, yellow and green variations of the character (simpliest way is to change hair color to blue, blonde or green respectively, you can also try using red, blu, grn and yel tags).
Added more dataset for different outfits and headwear (like a maid or bunny suit, so it should be easier to costumize the character).
Increased net size (to make it a bit more detailed and handle larger dataset better, also means that the file size is bigger).