This is an omorashi lora. It understands a lot of concepts: peeing, wetting, desperation, pulling clothing down or to the side, toilet use, different poses, different angles. Also check out the <a target="_blank" rel="ugc" href="https://civarchive.com/models/273003">SDXL version</a>, which is much better at conceptual understanding.Here are the relevant tags it was trained on:Core concepts<ul><li>peeing</li><li>wetting self</li><li>desperation</li><li>leaking<ul><li>Like "wetting self" but a lot less. Might need to experiment with the strength, or even add "wetting self" as a negative with low strength.</li></ul></li></ul>Pose/expression<ul><li>holding crotch</li><li>standing</li><li>sitting</li><li>squatting</li><li>lying</li><li>kneeling</li><li>relief</li></ul>Clothing<ul><li>panties</li><li>pants</li><li>shorts</li><li>leggings</li><li>swimsuit</li><li>clothing aside</li><li>clothing down</li><li>clothing around one leg</li><li>ripped leggings<ul><li>I only had a handful of images with this, it may not have learned it well.</li></ul></li></ul>Container<ul><li>toilet use<ul><li>you can use this one on its own</li></ul></li><li>peeing in container<ul><li>combine with a container from below</li></ul></li><li>holding container<ul><li>Use if the girl should be holding the container. It may struggle a bit with this, there's only so many training images of this concept.</li></ul></li><li>cup</li><li>bottle</li><li>teacup</li><li>bag</li><li>bucket</li><li>potty</li></ul>Angle<ul><li>from above</li><li>from below</li><li>from side</li><li>from behind</li></ul>Some of these tags are "compositional". E.g. desperation is a general look of needing to pee, holding crotch means actually holding oneself. If you want the typical "potty dance" pose you need both tags. Similarly, "peeing in container" for the overall concept, then "cup", "bucket", etc for the type of container.<h3 id="heading-457">Generation tips</h3>The most common resolution bucket in the training images was 448 width x 576 height. I typically use that res, with hires fix, scale 2, using latent upscaler. This lora is bit strong; depending on the model you're using it on and your exact prompt, you may try lowering the strength to 0.9 or 0.8 for a more natural look. It works best on NAI, or models closely related to NAI.

Files with SHA256: adf5a7728a0d2e11935f610f3d2ab6a1ac14b98d268d7b1f322c046efe966a4b