General usage:Depicts scenes in which a giant woman is licking or about to lick shrunken people. Generally speaking the format goes:gtsLick, giantess &lt;pose&gt;, tiny {man/woman} &lt;pose&gt; BREAK &lt;other details about characters and setting&gt;Poses for the tiny person include normal things like standing and sitting, but also "in hand/hands" or "dangling in hand/hands", which is self-explanatory. "dangling" is a bit more about the hand position. The tag "holding_tiny" can help put the tiny individual in the giantess' hands. Use the example images for prompt guidance. I have had about a 75% success rate.<hr />Other notes:Was hoping that using 'small' instead of 'tiny' would control the relative size (in my training captions, small was for ~1-4ft while tiny was for &lt;1 ft), but limited success on that. May have to try training a dedicated LoRA for that.Sometimes if the tiny is rendered small, they can lose a lot of detail (a common issue in generating giantess images). I've been able to reliably inpaint to get that detail back. 0.4 denoising, use "only masked" with some padding as the inpaint area to zoom in on the face with full resolution. I did that with several of the images in the example set.<hr />I struggled for over a month to train this on Pony and wish I had tried Illustrious sooner. In the pony versions, I could get some good images, but it was extremely difficult to get it to obey any specific prompts beyond the characters and setting. I wonder if it's either because of just how large and diverse Pony's training set is (and how much you're often fighting against its existing training without realizing it) or if it's the text encoder getting a little overwhelmed with all of the quality prompts.

Files with SHA256: 34f4d3d4e7a2d9d13cb39e7ada3544bcf9a15e9c7171b2ec05bb91f5f9def508