V2 illustrious:

-Supports more dynamic poses at the cost of angles (will be finetuning the tags/dataset for this).

-Can be styled better

-Results are better with full body

-Interferes less with other loras

-Trained on more images (spent my university nights to find them, support me (︶︹︺) )

Trigger word:

legsup_tj, thigh sex, lying on back, penis

Negative prompt (suggestion):

vaginal, vaginal sex, username, watermark, artist name, censored, mosaic censoring

Tips:

Diffusion models are weak at understanding:

• occlusion

• depth hierarchy

• object blocking another object

So you need to explicit occlusion phrasing, not just positional phrasing (e.x "penis between thighs" will not work as intended because the model interprets it as it being positioned in that region, not that it is occluded from view).

**************************************************
If you feel like doing so, please post generated images to this LORA's gallery (do so by directly clicking on the "Add post" button underneath this LORA's page and not cross-post it) so that I can learn from the prompts & results and adjust the training settings as I'm still a beginner in LORA training. As a compensation I will tip the best images.

P.S: I welcome suggestions/(advice) tips!

**************************************************

V1:

Better results with only "disembodied penis".

LORA strength: 0.6 - 1 (1 might be too strong)

CFG: 5 - 7 (5 usually gives good results)

Trigger word:

legsup_tj

Guide words:

on back, lying, legs up, thigh sex, penis

Negative prompt (suggestion):

vaginal, vaginal sex, username, watermark, artist name, censored, mosaic censoring

General Tips:
-> If feet look messed up, try adding "feet up" or "feet/shoe soles"