New Illustrious and Pony released!
SDXL version was a test to see how my dataset translates. May give poor results. After testing the SDXL version it seems to not work very well, but feel free to play around with it yourself, might be possible to get something out of it.
With this remake I decided to start from scratch, should be a lot more customisable and more compatible with other LoRAs.
This is a pov LoRA for hug and kissing with customisation prompts
Core Prompts
pov kiss default, 1girl, 1boy, pov,
pov kiss breast press, breast press, 1girl, 1boy, pov,
Customisation prompts
kiss,
french kiss,
close up,
from below,
from above,
dutch angle,
looking at viewer,
reaching for viewer, arms up,
imminent sex,
sex,
cumming, creampie, ejaculation, trembling,
pussy,
nipples,
penis,
fully clothed,
open clothes,
naked,
breast grab,
torso grab,
thigh grab,
arm grab
Description
This is a pov LoRA for hug and kissing in a sitting position with customisation prompts
Core Prompt (Use this if having trouble with consistency)
1girl, 1boy, breasts, pov,
Trained prompts
before hug,
Hug,
before kiss,
before french kiss,
french kiss,
kiss,
Helpful prompts (ongoing, might add to this in the future)
(breast grab:1.3),
(leg grab:1.3),
(torso grab:1.3),
(Head grab:1.3),
penis, vagina, vagina on show, nipples,
sex,
(Imminent penetration:1.3)
(Twitching:1.5), cumming, creampie,
FAQ
Comments (11)
Does the position well but affects the style a lot because i assume entire dataset is done in same artstyle.
Hey there, sorry you're having trouble with it yes the data set I generated for all my LoRAs are all in anime style not sure how well it will translate to anything else.
Was hoping it would still be possible with different checkpoints and the right mix of LoRAs it to get different styles, will experiment with it one day myself as the Idea of generating double/triple more pictures for the LoRA is quite daunting but I will try to do it one day if it's limiting the style that much
@Oscarc oh no i am also generating anime style. Issue is, since every picture in dataset has a certain style, generations tend to have same style of lips, eyes and background even at lower weights. As you can see from example pictures people posted. Issue is not the amount of pictures you have but most of them having samey faces.
Apologies if you intended this as a pose+style LORA. If that is the case it works perfectly. Just wanted to give you feedback in case style overcooking was unintended.
@Wizzi Ah ok that's fair, never tried changing the style when getting training data. Will try to include more styles in the next LoRA I make to see how it goes. Would rather not limit what people can make so hopefully this will work in the future, thanks for the feedback!
What style did you use in preview images?
@Rarn1234 Hey there, the checkpoint used was https://civitai.com/models/257749?modelVersionId=290640 and the style came from the character LoRA I used https://civitai.com/models/7715?modelVersionId=818180 might be able to use it for different characters if you don't include the trigger words from the character LoRA. Might also be worth copying the negative prompts from one of the preview images as that can effect the style also. Hope this helps
pretty good for anime/cartoon style
but doenst work really well to semi/realism which is too bad since i really like this LoRa but i prefer semi-realism
thanks
im not cryin ur cryin
Can you make a non anime version please?
Have been thinking about remaking this LoRA, so it's very likely the next one I work on! 😁 With the settings I use now hopefully the remake will be a lot more flexible.