Introduction
This LyCORIS is used to generate autopaizufella gesture.
IMPORTANT: PLEASE DON'T TRY PROMPT LIKE "BACK_VIEW" TO GENERATE ANY FUTANARI BACK TO YOU. YOU WILL GET CHAOS.
I MUST remind anyone who want to use this lycoris. It is just a lyco, not controlnet. You need to try some times to get what you want. So I recommend you use highres fix. If you have some good generations with any specific seeds, using highres to upscale the images to get better result. This lycoris works in 1024X1024 images, but bad in 512X512.
BAD IN 512X512.
How to use it
I recommand using it in weight 0.6-0.7. If you have other loras used at the same time. Bring it to 0.7-0.8 or higher. Weight 1 is also acceptble. But you may find the images get blurry at some times.
This lora has two triggerwords. One is autopaizufella, use to generate penis without cum. The other is autopaizufella_ejaculation. Use it when you need cum in your image.
Training details:
Traingset is of about 400 images. They are mirrored before training. Half of them are chosen as regularization images. Total steps are about 3840.
Regularization = true
resolution=512
batch_size=2
epoch=10
network_dim=32
network_alpha=32
clip_skip=2
Using AdamW8bits as optimizer:
lr="1e-4"
unet_lr="1e-4"
text_encoder_lr="1e-5"
Locon parameters:
conv_dim=4
conv_alpha=4
Multires technique parameter:
multires_noise_iterations="6"
multires_noise_discount=0.3
Description
FAQ
Comments (3)
The most common problem encountered was that the penis, instead of going from base to mouth, would morph into her belly and reappear between the breasts into her mouth. From each batch I made, I only got around 3 or 4 per batch that were good, yet some didn't even perform the paizuri or fellatio action. Also, the model seems to have a preference for white hair and cat ears, since I never really added tags related to hairstyle or strong characteristics.
For the penis problem, yes a lot of trials are needed. But I noticed the "fellatio" and "between tits" prompt in your image. They may be the problem. These concept in SD models are mainly trained based on the normal sex position. Even after applying this lora, these concepts cannot be completely replaced with the futa ones. So delete these prompt related to normal sex, leave only the "autopaizufella" may do the trick.
For the variation problem, obviously the lack of dataset. I am working on a bigger dataset trying to solve this problem.
@DeepDark_Fantasy514 Hello! I actually found a more consistent set of tags with just your specific tag and 'fellatio'. I uploaded some of the ones I got, with a more success rate of 3 every batch of 4.
Also, I apologize in case my original comment appeared too judgmental. I'm very new to AI Generation and wouldn't like too be too harsh without knowing more of what I'm doing.
Now that I have a better understanding, I see that it's a very amazing work!
Thanks again for the advice and the model.