V2 (Illustrious/NoobAI) Information:
This one is meant to be used at 1.0 strength. No more, no less. (Not compulsory, feel free to experiment).
The trigger words are also a bit different this time around, and having the man in the frame can be kind of difficult. Refer to the prompts in the showcase images for a better idea on how to correctly prompt for different situations. Some of the information in the v1 section may also be relevant for this one.
All showcase images are made on tensor with 25 steps initial, 25 steps hires with latent upscale at 0.4 denoise, and adetailer for the faces. They also use SGM_uniform with DPM++ SDE. A latent upscale hires fix does wonders for the gen, and I find that only possible to do online on Tensor. If you want to run it there, here is the LINK.
head on lap, penis in mouth, oral, fellatio, from side,Charging a very little amount for buzz per download for a very limited time to fund development of Illustrious versions of my other LoRAs.
Models like Pasanctuary and AAA break some LoRAs. This is one of them (not thoroughly tested). Use WAI-NSFW-Illustrious for online civit gens. For local gens, I recommend Illustrious Pencil, I love the details in that one.
V1 (PONY) Information:
For this one, try not going over 0.95 strength. The LoRA supports both 1216 x 832 and 1344 x 768. Don't go above or below this resolution. 0.9 is the optimal strength to use.
If you want to have the man's face in the image you gen, just include "staring, smiling," after 1man in your prompt (or just refer to the gallery images with male faces in them). Please follow the prompting method below to get better gens. Also, portrait resolutions don't work.
NOTE: If you choose to include the man's face in your gen, it may bleed facial features from the girl's face. Some inpainting should fix this.
Follow this prompting style for best results (checked with minimal testing) (Mostly relevant for v1 only):
First Part (LoRA trigger): 1man, face out of frame (OPTIONAL), sitting, girl on lap, penis in mouth, bl0wjob,1girl, head on lap,
Second Part (Example description of girl): 1girl, snow, pointy ears, twintails, green eyes, outdoors, elf, long hair, white capelet, solo, winter, white hair, hand on ass (OPTIONAL), lap lying blowjob,
Third Part (Background details and other stuff): outdoors, very dark night, purple clouds, night, on sofa, balcony, dark, lake, aesthetic, beautiful
If you are using A1111/Forge, and the LoRA does not show up, you can force it to appear by enabling the "Always show all networks on the Lora page" option on the settings page. (Thanks to @pr0log for pointing this out).
Usually, 0.9 strength works best, but if you are having problems with the style, lower to 0.85 (lowers chance of guaranteed pose). If the girl's body anatomy gets wonky, lower the character LoRA weight by 0.5 at a time, until it gets fixed.
If you want to support more models like this, or you just like what you see: ko-fi (｡♥‿♥｡)
--Reserved for Future Updates and Observations--
Description
FAQ
Comments (14)
my only complaint is it's hard to get the guy's face in frame
EDIT: just include "staring, smiling," after 1man in your prompt. Seems to work most of the time.
Yeah, it was trained on half out of frame male faces. Maybe I'll release a version with the faces in frame. Also thanks a lot for the buzz.
Maybe try 1216 x 832 with more male hair descriptions in the prompt with the LoRA at 0.9? I'll have to test.
Also, not tested this LoRA, but about to.
But I found with pony sometimes adding a minor description of the man will add face.
e.g. "man with short brown hair". Worth testing I guess?
One of the most consistent LoRas I've ever used. Have some buzz my friend.
Posting some tests now.
Cheers.
Thanks a lot for the buzz!
@nottheninja Np mate... just posted a bunch of tests... works nearly every time.
one of the most consistent LoRAs I've messed with. LOL
Don't know how you did it, but with the exception of things like epinoise LoRA for SD1.5 this basically works like 95%+ of the time. Even with other LoRAs as I tested. Good work mate.
@NyxxiNyx Glad you liked it! It all depends on the quality of the dataset imo. I have many more ideas that I plan to release LoRAs of. Stay tuned!
It also takes good prompting to make sure the LoRA works well. Your results are exceptional, and even I could not replicate that level of consistency when doing tests with this LoRA. I'd love it if you could share some of the prompts for me to try out later.
@nottheninja I'll DM them now. Also liked your Previous LoRA. Was fire mate. thumbsup
This one doesn't show in my lora list, thats a first for me. But still works great
If you are using A1111/Forge, and the LoRA does not show up, you can force it to appear by enabling the "Always show all networks on the Lora page" option on the settings page.
Can this work in Diffusers? Getting this error: .local/lib/python3.10/site-packages/diffusers/loaders/lora.py", line 298, in lora_state_dict
state_dict = maybemap_sgm_blocks_to_diffusers(state_dict, unet_config)
File ".local/lib/python3.10/site-packages/diffusers/loaders/lora_conversion_utils.py", line 57, in maybemap_sgm_blocks_to_diffusers
I unfortunately haven't tested the diffusers pipeline yet, and won't really be able to help with this. I mainly use ComfyUI, where the LoRA works pretty normally. I also had some users complain about how the LoRA does not show up in A1111/Forge. Maybe it's related to that?
This LoRA was trained on the Pixai trainer, which does cause issues like this. I plan to train future LoRAs with Kohya (on the civitAI LoRA trainer).
Favorite lora I've downloaded in a long time. Tried my hardest to prompt this with no success a few weeks ago and was happy to see it had some variation with the girl being slightly behind, to the side, or even showing ass to viewer.
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.