Gives dudes a big raging hard-on in their pants.
In the prompt, use the trigger word "vpl" and maybe what kind of pants/underwear the subject is wearing (so far underwear works best.) Throwing the word "bulge" in there may yield better results under certain conditions. It seems to do better with semi-realistic models. The weight is most effective around 0.7-1.0.
P.S. this is my very first LoRA, and I intend to release an improved version in the future, so stay tuned! If you've generated any good VPLs, please post or share as I would love to add it to the training data for v2.
Description
Trained on 139 images, 2 repeats for 50 epochs. Saved and sampled the model for every epoch completed. Epoch #22 out of 50.
[[subsets]]
num_repeats = 2
keep_tokens = 1
caption_extension = ".txt"
shuffle_caption = true
flip_aug = false
is_reg = false
image_dir = "E:/Projects/vpl_lora/v7/dataset"
[noise_args]
[logging_args]
[general_args.args]
pretrained_model_name_or_path = "E:/Projects/v1-5-pruned.safetensors"
mixed_precision = "fp16"
seed = 23
clip_skip = 1
xformers = true
max_data_loader_n_workers = 1
persistent_data_loader_workers = true
max_token_length = 225
prior_loss_weight = 1.0
cache_latents = true
max_train_epochs = 50
[general_args.dataset_args]
resolution = 512
batch_size = 2
[network_args.args]
network_dim = 16
network_alpha = 9.0
[optimizer_args.args]
optimizer_type = "AdamW8bit"
lr_scheduler = "cosine"
learning_rate = 0.0001
[saving_args.args]
output_dir = "E:/Projects/vpl_lora/v7/build"
save_precision = "fp16"
save_model_as = "safetensors"
output_name = "hardvpl"
save_every_n_epochs = 1
[bucket_args.dataset_args]
enable_bucket = true
min_bucket_reso = 256
max_bucket_reso = 1024
bucket_reso_steps = 64
[sample_args.args]
sample_prompts = "E:/Projects/vpl_lora/v7/prompt.txt"
sample_sampler = "ddim"
sample_every_n_epochs = 1
[optimizer_args.args.optimizer_args]
weight_decay = "0.1"
betas = "0.9,0.99"
FAQ
Comments (7)
omg yes we needed this!! thank you
What was your dataset like? I’ve been trying to improve the outcome of unzipped pants but not sure how to describe it when training.
The training dataset doesn't have any unzipped pants, apologies. The LoRA is trained on visibly hard erections in underwear/pants/jeans for simplicity and consistency's sake (it's shocking how many "kinds" of bulges there are.) However, you could try to prompt with "pants pulled down" or "pants pulled down around ankles" along with "underwear" or "underwear tease". I've had success with that generating unzipped pants and a nice bulge that way through bulgerk-dickprint, but haven't tested with my LoRA (yet). I hope my answer gives you some guidance!
i actually had the same request...I attempted to train this concept and failed. would it be too much to ask you to share the dataset you used in the post. I would love to study it and see why my dataset failed.
@omegablast20023899 add me on discord: jakespillstea
@omegablast20023899 i suggest looking into masked LoRA training
also works surprisingly well for cameltoes in some models (galena etc.)
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.