hey everyone! This update was one I was always meaning to do but never got round to it, so I'm happy to finally get it done! I was commissioned for this update by Nyxy_style https://civarchive.com/user/Nyxy_style, not something I normally do as this is more of a hobby for me and I've never really thought about taking it further, think of this as a test drive of sorts.
This is a pov sex position Lora to have the character lean against a wall. With this update, the LoRA should be a lot more friendly with other LoRAs and be very customizable, so please feel free to play around with prompts and see what you get! 😁
core prompt
poagwasex, standing, pov, 1girl, 1boy, pressed up against wall, wall, leaning on wall,
customization prompts
from above,
dutch angle,
kabedon,
arms crossed above head,
arms up,
(standing split), one leg up, leg grab, feet,
imminent sex,
sex,
after sex,
sex from behind, ass,
paizuri,
fellatio,
handjob,
penis,
pussy,
anus,
(cum in pussy, overflow, cum bubbling:1.1),
(cum on body, cumshot, cum bubbling:1.1),
arm grab,
breast grab,
ass grab,
Description
Experimental version
Never used SD 1.5 before so might not work as intended. Depending on how this goes I might try to make SD 1.5 versions of my other LoRAs too
This is a pov sex position Lora to have the character lean against a wall
SFW prompts
Pov against the wall, no sex
Pov against the wall, no sex, people in the background
NSFW prompts
Pov against the wall, one leg up
Pov against the wall, standing
Pov against the wall, standing doggy
Pov against the wall, straddling man
Prompts to use with the above
people in the background
creampie
FAQ
Comments (12)
Great job.
Any idea how to make the female breasts pressed up against the male's chest? Been trying to create this with this LoRa, but overall great stuff!
Not easy that especially because the training data does the opposite. Might try updating this LoRA to include it in the future. In the meantime I had some success on lower strength not perfect though. https://civitai.com/posts/3745066 here's the link to the post so you can try remixing. Hope you have some success! 😁
@Oscarc Yes! I'll give this a try, thanks!!
@ToastedSub There's also a lora for this, one of the breast press ones. Perhaps if you combine them you can get what you're looking for.
A little confused.. The grab prompts aren't in the training data? Does that mean there were no images of it or they weren't tagged? Why aren't they the usual danbooru tags "grabbing another's breast"? Why is only the word "grab" weighted up and not the whole thing?
Hey, there are no images of grabbing in the training data. This was the 1st LoRA I tried this and it worked well so it's how I train all my LoRAs now.
The grab trigger words are how I usually get grabs to work when generating, so they are only there as a baseline to help people get started. If you know of a better grabbing prompt feel free to use them should still work fine I hope. Hope this clears things up 🙂
@Oscarc Ah okay, the bit about training data makes more sense. It seems to work well.
It's not the specific words you've chosen to use that are confusing, it's your weighting method. "breast grab:1.3" is only weighting the word "grab".
To weight multiple words they need to be in brackets "(breast grab:1.3)" will weight both words. In Auto1111 and I believe in Comfy you can just select the words and use Ctrl+Up/Down.
In InvokeAI this would be (breast grab)1.3
@blisterfingers thank you so much for pointing this out! I only generate through the website and been winging it so I don't know these things. This will help me big time when prompting to generate training data for LoRAs. Will go through and change the trigger words now. Thanks again!
@Oscarc No problem. With what you said about the training, I just wondered if there was some specific reason for it. Nice lora, though. Definitely gonna use it. ♥
@blisterfingers the main reason was because when I had grab training data for a LoRA before this one, it caused hands to have way to many fingers to the point where 10 fingers on one hand would happen sometimes. So that's why when I made this LoRA I only generated training data for the poses only, to see how well it performed.
@Oscarc Interesting. I'll have to keep that in mind.