This model was trained on 3-4k images where hoses are inserted in the mouth or butt. Naturally it has lots of inflation related content in the dataset, but the main purpose is to make hose insertion more reliable in SDXL models.
This dataset was extracted from the Hyperfusion dataset, so the tagging will be similar. Primarily use "hose in mouth", "hose in butt", or "holding hose"
Other Inflation related tags may also work, but I haven't really tried.
Description
Training Notes:
LR 2e-4
frozen text encoder
dim 16
alpha 8
conv_dim 8
conv_alpha 4
batch 4
GA 8
4k images
network dropout 0.3
tag dropout 0.1
scale_weight_norms 1
ip_noise_gamma 0.02
min_snr_gamma 1 (its been suggested that latent diffusion models be set to 1 or 2)
soft_min_snr
keep_tokens 1 (for the primary 3 tags)