Not recommended for realistic model
Main tag : lactation, projectile lactation, lactation through clothes,
Secondary tag : upper body, nipples,
LoRA str : 0.8-1.2
Prompt str : 1.0-1.2
Denoising str : 0.3 to 0.45
Training Powershell info :
Training dataset Image : 460
DIM/Alpha : 16/8
Train batch size : 2
resolution : 768
mix/save_precision = BF16
optimizer_type = AdamW
Huber_c = 0.40 [SNR]
Multires , Ip_noise , Noise_offset , Min_snr = Disable
GPU : RTX 4060 16GB 160W
Torch 2.5 + cu12.1 + Xformers + gradient_checkpointing
Vram used : 10.1GB
Speed : 1.60s/it
Description
This version adds more steps, sometimes overfitting if not enough prompts
There are 3 differences in the form
lactating into cup, lactation through clothes, lactation
FAQ
Comments (9)
You should put "milk" in your tags since that is what a lot of people will search on.
oh thank you
I have a similar issue with this model as the milking machine one - it has the inability to adapt to a given prompt.
pretty much yeah. It does what it does really well, the rest not really
Yeah sometimes if there's a very few prompts, it can overwhelm things if the weight is high, but the more prompts you put out to try and steer it to a different thing, the more weight you need to add to get it to pop into the situation... if it ignores many of the prompts, then reel back a little... it's quite a balancing act but sometimes, when you find just the right balance, you can get things aligned into the same image. Or at least close, then I usually fix the remaining oddities with inpainting, sometimes with all new prompts and run until satisfactory.
...and artists say this takes no work. ;D
(take the last comment with a grain of salt, it's intended very tongue-in-cheek!)
I can use lora with other models? Instead of anything model? Thanks!
I tried it with ExpMix_Line, and it works fine.
What noise do you have for your ancestral samplers? I don't quite get the same results.
On that note, how can I put the image in the comment?
kyonyuu fantasy cg?