please keep it clean and respectful while using this lora

V2.0

balanced the three types of penis in dataset, and add cum conception into the Lora

How to use:

3 types of penis are supported:

Upward, forward and Downward penis.

use big/small to adjust penis size
(fx. small forward penis)

Liquid interaction is supported:

liquid (milky cum/beer) ejaculates/drips from the tip of his penis, falls on/into location/objective/container.

available both in T2I and I2I

works on Futa as well !!!
your support and comments are my motivation

V1.6

use close-up penis image and image pairs (penis & no penis) to train this model.
able to add three types of penis (upward, downward, forward)

next version will try to add conception of cumming in this model.

V1.5 (b):

used AI generated beautiful cook as training data, the penis looks more tempting.
It works for T2I and image edit. This one can always generate penis, but the type of penis is not very controllable, further training will come in the future.



However, this version works very good in transforming animated penis to realistic style.

use prompt: '将平面插画转绘为真实摄影: hairy, muscular naked male with a penis in his crotch. Milky cum ejaculates/drips from the tip of penis.'

trigger word:
a forward penis in his crotch.
an upward penis in his crotch.

a downward penis in his crotch.

may work:
a downward penis in his crotch.

Milky cum drips from the tip of his penis.

V1.5 (a):
used AI generated beautiful cook as training data, the penis looks more tempting.
It works for T2I and image edit. works good for upward/downward erect penis, but may fail sometimes

trigger word:
a forward penis in his crotch.
an upward penis in his crotch.

may work:
a downward penis in his crotch.

Milky cum drips from the tip of his penis.

V1:
Prompt:

erect penis
show his anus, and erect penis

dataset:
Trained on 26 image medium-shot + close-shot (naked)
firstly use Klein 9b to put clothes/shorts on those images (clothed)
caption the clothed image in Gemini, require it not describe person's appearance, and then manually deleted all 'pants/shorts/underwear wearing' description. use those captions for naked image set

training:

trained in I2I mode by Aitoolkit, using clothed images as input, naked images as target.

Trained with 2e-4 learning rate, 32 rank for 2600 steps.
training resolution:521 and 768
turn on Differential Guidance and set it to 4 instead of 3.

other setting was in default.