<p>Concept LoRA for women peeing.</p><p></p><p>Trained on 2000 photos + 800 "supplementary" anime images. The photos have both a caption version and a tag version, trained 50/50, so the lora works with both prompting styles. The anime images were tags only, and used only to help emphasize certain concepts underrepresented in the photos. You can force it to do anime, but it's not great at it; the model is intended for realism.</p><h2 id="generation-settings-nz9gof88p">Generation settings</h2><p>You <strong>MUST</strong> use true CFG to get the best prompt adherence. But you still need to use one of the antiburn techniques, e.g. Dynamic Thresholding, AutomaticCFG, or SkimmedCFG. I used AutomaticCFG for all the showcase images. Lower CFG is generally more realistic and less oversaturated/burned, but worse prompt adherence and worse anatomy.</p><p>Guidance Scale: 3-10</p><p>CFG: 2-6</p><p>TBH I don't really know, using flux with an extensively trained lora like this one is kind of weird. It partially loses the guidance distillation in the model, so you have to use CFG, but you still have to use an antiburner. Just play around with things.</p><h2 id="prompting-guide-pbh0fkpji">Prompting guide</h2><p>TLDR just describe what you want in natural language.</p><p>To generate the captions, I used multiple VLM models, then fed those captions + the image tags into an LLM and had it combine everything into a final caption. So basically any way you can think of to express the below tags in natural language will work.</p><p>Make sure to mention "pussy" somewhere in the prompt if the woman's pussy should be visible. The model is quite responsive to this word.</p><p>For photos trained with tags, the caption was "A photo of a woman.", "A photo of two women.", etc, and then the list of tags. For anime images, the caption was "An anime image.", and then the tags.</p><h2 id="tags-bhzolrm82">Tags</h2><p>If you want to use tag style prompting, or get an idea of what the model understands, read on.</p><h3 id="core-concepts-vdvnf7jmj">Core concepts</h3><p>peeing, wetting self, desperation</p><p>peeing in toilet, peeing in urinal, peeing in urinal backwards</p><p>peeing in cup, peeing in bottle, peeing in bucket, etc</p><p>You can say "forcefully peeing" or "heavily wetting self" to try to make more pee.</p><h3 id="clothing-position-boltsa5yv">Clothing position</h3><p>pants down, jeans down, leggings down, etc</p><p>panties aside, bikini aside, etc</p><p>panties around one leg</p><h3 id="pose-s6cpynh46">Pose</h3><p>squatting, crouching, standing, sitting, kneeling, lying, knees up, leg up, legs together, legs apart, arm support, leaning, bent over, facing away</p><h3 id="expression-9x2wxfhz0">Expression</h3><p>smile, parted lips, open mouth, closed eyes, relief, embarrassed, looking at viewer</p><h3 id="angle-u6lzsrrie">Angle</h3><p>from above, from below, from side, from behind</p><h3 id="other-pq0mte6z3">Other</h3><p>outdoors, indoors, nature, urban, candid, night, over edge</p><h3 id="quality-tags-4a1mvcz1d">Quality tags</h3><p>rating_1 is worse quality, rating_2 is medium quality, and rating_3 is best quality</p><p>source:video and source:4k_video for images that were screencaps from videos.</p>

Files with SHA256: da2bbbfc1711ca309a14698e7567c04f64c6284d17e9fea78984370609d03ebf