Usage
Trigger word is 5tar5, strength at 0.6 - 1.0.
Check the model details, but there are a wide variety of your favorite positions and situations accounted for -- try them out!
About
Lets try something new! This model is trained on a selection of content from a specific JAV studio.
This was only meant to be a proof-of-concept, as the dataset was somewhat limited, but the results were so good that I thought I'd share it right away! More improvements will be coming in the future.
If you try it, I'd love to get your reviews / comments!
Have fun!
Details for nerds
Trained on ~800 images
mixed_precision = "fp16"
clip_skip = 1
max_token_length = 150
epochs = 20
repeats = 4
resolution = 1024
batch_size = 2
network_dim = 32
network_alpha = 16.0
optimizer_type = "AdamW8bit"
lr_scheduler = "cosine_with_restarts"
learning_rate = 0.0002
Description
FAQ
Comments (2)
It's odd. Without the pixelated parts, I feel nothing.
This Lora was trained using only uncensored images. I have a bunch of censored content that I'll try out soon!
Feel free to add "mosaic censoring" to the positive prompt though. Whatever floats your boat!