A lora for hiding your sword from the police
throatfucking
trained on 97 images
Image info
Model used abyssorange2hard
Model used based64
Notes(Update)
v2.0 is a LORA NOT A LYCROIS
works much better and much more consistent
much more flexible( prompt for the girl to be kneeling/squatting/sitting etc)
Redid the entire dataset (available for download)
Training settings?
The most important part of training is getting a good dataset!
Training settings can be found by installing the additional networks extension in auto1111
Extensions -> available -> load from -> additional networks
Restart webui
place lora into stable-diffusion-webui\extensions\sd-webui-additional-networks\models\lora
To view the training settings
Go to additional network tab -> choose model -> training paramters
Description
-Remade the entire dataset (available for download)
-changed from lycrois to lora
-MUCH MUCH more flexible
-fixed spelling mistakes
-fixed combability with other loras/models
FAQ
Comments (19)
Awesome update, boss~! Works exactly as advertised. Had to add "sword" to the negative for my model, but mostly because I'd bumped the strength down for flexibility. Any chance for total visibility adjustments that could include the full view of the back of the standing partner? Icing on the cake would be front (back? 180 turn, ya get it.) view too, but I know it's a big ask.
Either way, I hope you appreciated the process. The result was worth the effort IMO.
thanks for your feedback! As for your questions, I had some issues making the full body visible as most of the time it would just end up with horrible anatomy, hence why v1.0 had some issues so I removed them completely in v2.0 I intend to gather better images for it and retrain it in the future. For the 180 degree view, Im currently training a separate lora for it so it should be up soon
@mnemosynekotorin did you upload the 180 degree view?
btw kotosmix is one of my favorite model, great work there.
I cant get anything as anatomically correct as the reference images despite copying and pasting the exact prompts
that's a bit odd but try and following these steps to make sure you dont have any issues on your end
1-Did you make sure to place the file into the LORA folder and not the LYCROIS folder (due to v2.0 being a lora and NOT lycrois)
2-are your negatives needlessly complex and too long (if so, reduce them)
3-do you have the negative embeddings that I used
4-Is the lora correctly being recalled either through additional networks or the base gui
Just dont work. I do my best
Just dont work. I do my best
Just dont work. I do my best
I have tried everything with this in Graviti, and SOMETIMES I get a usable picture, most of the time it's just a bunch of swords, or no male at all. If I generate from the same exact picture that turned out good, tags, Seed, steps, I'll STILL just get actual swords. I have NO clue what I'm doing wrong. I've followed everyone's advice and I'm still messing it up. I wish "sword" wasn't in the trigger word
then its most likely a problem with your model or the lora isnt being called correctly take this example https://files.catbox.moe/t21vin.png its a VERY Basic prompt and it still works and i've yet to generate a sword which is something I'm trying to do and I've tested more than 20 models. If your stil having issues then tell me which model you've used and ill see if theres a fix for it
@mnemosynekotorin I think I figured it out, it was probably all my fault for being new and not knowing how to use these properly. Also, it's clear that most of these were made with hentai in mind and here I am trying to make furry porn with it. I did get some good results though, once I figured out where I was messing up
@reptilian321616 and what did you do wrong? i have problems aswell and am very interested in your solution
Where are people coming up with V3?
v3 was just the name I gave the model thats all. (we dont talk about v2) all the gens with suffix 3 come from v2.0 on civitai. Though it would probably be better to rename it to v3 but that comes with issues when people ask where v2.0 is lol. so in short v2.0 = v3.0
ive gotten really close to the source image but just some minor detail is off and i have no idea why. image for reference
does this have a problem with character Lora's? I just can't get it to work with a character I want
you might have to lower the character loras strength.
since the character lora will want to show the characters face and the sword swallowing position does not it will create a conflict
@mnemosynekotorin I've finally got it working, this tip was the best
Stable difusion can't detect it :(
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.