Do you want to support me? Consider being a patron on Patreon to vote on train polls and access stuff early. ❤️
Trained with over 400 images.
Doesn't have a specific trigger word general prompts for the finger in another's mouth concept
finger in another's mouth,open mouth , tongue, teeth,tongue out, pov
It also works really good with my Cum on Tongue LoRA and I highly suggest it :D
It also works with sex positions but might need some rng on hands or the position in general.
Also It doesn't have to be NSFW use only you can create funny SFW stuff with this lora
This LoRA enhances Finger in mouth concept
Discord server feel free to drop and say hi ^^
Description
FAQ
Comments (14)
Your Loras with poses and act are great! Keep the good work! It's so much usefull than 100500 characters and model mixes.
Thanks. I train characters too though :D
Sometime it is hard to find pose or concept to train ^^'
@Murky there are always tons of poses to train. :-) Even those that are quite simple help to make character do what you need without NASA prompt engineering. :-D
does it work with other loras?
Yep
@Murky awesome
Another banger. i really need to keep suggesting ball squeeze lora sir. Please consider ! :D
Love the stuff you've been putting together!
Question: what's you're workflow like?
I've got a few concept models I posted, and on my 4090 they've all taken four to five hours to train for each version. That's at 768 with batch size of 4.
Is it similar for you?
Characters I can get in 10 minutes, but concepts seem to take forever to train, in addition to the significant amount of time it takes to make sure everything is tagged and organized right.
And even then. I've tossed at least two bad models for each of the decent ones I get.
0.0001 lr
10 epoch
10 repeat
Around 150 200 for a decent concept. Nothing too complicated
@Murky I think I'll try that, thanks. I've been doing much larger numbers.
20 epochs, and then load balancing the repeats to get 1k total for each concept I'm trying to train into the lora, the penis_on_face one in particular had like 12 separate concepts in it, at 1k each.
I really wanted the cheek smushing affect, and without training a significant number of sfw, and counter examples, I couldn't get that working.
@Murky Do you shuffle your captions?
@aurenigma yes shuffle on. Large number dont mean good all the time. I keep raw training steps around 20k so with batch size thats 5k for you
And how it calculates is
Image x repeat x epoch
High repeat better for character and poses meanwhile high epoch for styles
Just hope for an SDXL version of that Lora, trained on better resolution images soon ! :) I'm playing with PonyXL and XL's understanding of tons of new concepts is really outstanding.
You're a legend, thanks.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.