I've created a Lora specifically designed to capture poses related to the themes 'HAVE TO PEE,' 'HANDS BETWEEN LEGS,' 'HANDS BETWEEN THIGHS,' 'KNEES TOGETHER FEET APART,' and 'COVERING CROTCH.' These are particularly challenging poses to depict accurately, and while there are existing Loras with similar themes, none focused solely on these poses.
This new Lora enables the creation of two distinct scenes: one where the subject is visibly struggling and holding themselves, and another that zooms in on the crotch area.
However, as you can see in the image, when I tried with ryuzaki umi and Sailor Mercury's Lora, I could not produce an image concentrated on the lower half of the body, and there is a phenomenon of difference depending on the Lora.
This is a selfish guess, but could it be that some Loras cannot generate images due to the lack of material for the lower half of the body? So, there is a possibility that you cannot generate a covering crotch image of your favorite character even after trial and error.
Description
FAQ
Comments (5)
I guess after all that excitement, picked up by a giant eagle monster, almost being strangled by a sea snake, falling off a waterfall, getting stuck in a sand pit and gears and warping through space and time, I'd have to go pee really badly too
And you didn't mention the tentacles... twice!
@Gaspatcher I'm a man of standards
I'm not expert about LORA, but on my experience as user, I noticed there are LORA born to add some small details, like a pair of shoes, and LORA attempting to work on entire image, like costumes or poses. In this last case, the LORA can be bad trained, since it was not able to separate the concept and tends to drive also the face and the body of the charater. You can see this is you set a second LORA with a specific character, the result may work but the quality looks 'stressed'. With anime this may be more difficult to notice, because they use simple colors and lines. The fact your LORA cant visualize the lower part may be (?) because it tends to drive the image to the dataset, instead of separate the concept.
Anyway I can assure SD is capable to invent views and details even if they were not in the dataset.
Now this is awesome!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.