lora trained on a dataset of recent thepit artwork. Allows to make bimbo women in the style of thepit as well as some nsfw, such as blowjob and bukkake.
Description
This new version is a bit better for mixing with my other bimbo loras. The overall style stay also more consistent compared to the v2.
FAQ
Comments (22)
Awesome ! Can you do the same with his old art ?
Sorry but there is no way I could make a lora as good as this one if I was using only his old work, because I would have much less data to train the model on. Maybe I will try at some point...
@Plankton_the_Great I understand, I knew it would be hard. Good luck if you try anyway
This is amazing! I've been searching for a 'John Persons' -style comics LoRa for a while. Couldn't find any. Like at all... I actually was surprised no-one seemed interested in it. So started creating my own (at little avail). Never thought of using other search terms like 'The Pit' and just stumbled onto yours minutes ago accidentally haha. Gonna give this a go. Thnx and keep up the good work!
I usually don't comment on anything ever, but this thing just deserves an endorsement. It is the best lora I have ever worked with. The flexibility is off the charts, it's a fucking style lora that you can mix with other style loras without making a mess. It also doesn't matter what you put this over as long as its a female face its gonna turn it into a bimbo without looking garbage. I wish other loras were this good.
Hey, probably the easiest way to contact you.
I was planing on doing a lora from the artist "lagomorph" and was wondering if you would share a bit of knowladge on how you created your "Pit" lora
Hi, I don't have any special trick but I can give you some advice.
1) Prepare the dataset by cropping the images so that they show what you want the lora to output. If an image shows multiple subject and the resolution is high enough, crop the image in multiple images, one per subject. Get a mix face of close-up, some upperbody portrait and full body.
2) If there is a lot of speech bubbles, try to remove as much as you can by painting over them with Gimp or some other software.
3) For captioning, I used WD1.4 and filtered out some keywords such as "large breasts", "thick lips", "makeup", etc.
Maybe I did not find the right learning parameters but in my experience I had to train for a lot of steps to get close to the style I wanted, more than 10 000, and then I finetuned the model even more with a smaller learning rate for a few more 1000s of steps.
@Plankton_the_Great
kk thanks,
the first batch was already okay, painting over the speech bubbles is a good idea and i'll try that with the next go and with more steps, i had now 4k.
How many pics would you say? my character loras i did with 140-240 pics, same range for a style lora?
also did you include landscape shots/pics without a subject?
As for finetuning i have to read into that so far never even heard of it :D
@caphalorthrow I did not include any image about landscape, only images where the main subject is a woman. I used about 500 images but you can probably get better results with less images and better captioning. I think 100 is the minimum if you want some flexibility though. Fine tuning is simply re-train a model already trained: instead of starting from scratch, you load the weights of a model trained previously, and continue training, with a lower learning rate for example.
@Plankton_the_Great
Allright then i need to find another 2/300 or so images for my training data set :D
And thanks for the pointer for the finetuning
also my discord, a few have made some great stuff with your lora
Very very good, but you need to be careful with the weighting while using it. I tend to like the images more with just a touch of the bimbo style, which this does a good job at.
What's surprising is that it appears to work well with a wider variety of models than I expected, from anime to photorealistic models.
I think it works well with all models simply because I trained this lora on the base sd 1.5 model. Thanks for the appreciation!
I love this Lora, but the checkpoints I use are SDXL, therefore we need this Lora in XL. I hope it will be soon.
With the new civitai training tool it is very simple.
You are amazing dude, I don't suppose you take commissions for Loras do you? I can't wrap my head around setting one up, would love to do sparrow's art
This is one of my all-time favorite Loras! Are you considering making an SDXL version?
Outstanding work.
Would it be possible to make a version that does make every penis black? I'm using this at my upscale stage and even on a low weight like 0.25 it redraws all male characters as black.
best lora maker for bimbo lovers
lol. It just so happens I was looking for something like this for help with a comic book I want to create. It was a match that was meant to be.
Please make an SDXL version this is one of the best loras out there
Is it possible to do sex scenes with this Lora?
I can't find that lora in generator images? It was deleted or what?