A test model for niche purposes, for giantess with snot wirh tinies

Giantess snot

Giantess_POV_Vore_Mk_VI.safetensors

v1.0

tl;dr:Makes a humanoid character look like they're about to eat you. Can include a wider mouth, an intense stare, a more dominant or malevolent expression, extended tongue, orienting the face to give you a view of the throat, rendering the arm to imply she is holding you, and sometimes improving rendering of mouth interiors. At extreme close-ups, can also give you a mawshot image. Flexible to a variety of characters, body types, ages, and skin tones. Playing with teeth and closed mouth can produce confusing sexy nightmare fuel. Sometimes the tongue gets a little out of hand in semi-realistic models.Long version:I had been struggling to find something for a POV vore mawshot that wasn't oriented around furries, and trying to get there by prompt usually ends up as a fellatio perspective shot, too far from the character's face to be interesting and not really oriented quite right. Fine-tuning of angles, poses, facial expressions, etc. can be a lot of trial without controlnet and even then it can be a pain. Facial expressions can also be rough for this kind of context given the bias of most of these training sets toward more 'submissive' facial expressions.The general effect of this LoRA is to make the character look a bit more "predatory" - a tiny bit more monstrous and intense and generally oriented to give you a direct line of sight into her mouth. It also can sometimes improve the rendering of mouth interiors. It can also provide a "mawshot" image if you prompt it for (extreme close-up:2) in a way that often turns into a mess without the LoRA (see the included grids). The results for those extreme-closeups aren't incredible, but it's a lot better than I've been able to get without the LoRA. I'm working on a separate LoRA to specialize in such images.I trained this exclusively on female characters and don't know how well it transfers to male or non-humanoid characters (though feel free to give her purple skin and fangs). I did endeavor to train it on a variety of body types, skin tones, ages, etc. and it seems to play nicely with prompts and LoRAs on that front. I also trained it exclusively on illustrations/hentai, though it seems to reliably do what it's supposed to for semi-realistic checkpoints as well, though the tongue and lower lip can get a little confused. I've generally had good luck with a strength of 0.8-1.0.Most of the images I'm including are straight 512x512 with no rescaling, no inpainting, etc. so you have an idea of how it performs right out the box. If I get the buzz to train later revisions, I'll likely improve the training on full body shots, get more control of the tongue shape, hopefully get more control over arm placement, give it more variety of backgrounds and lighting conditions, etc. I probably won't try to train it any better on extreme closeups, as I think that needs its own dedicated model.Some relevant tags:open mouth, tongue, saliva,drool,teeth, uvula are pretty standard. places nicely with a smile or evil grin"arm" or "inHand" - whether the pred's arm is placed to imply the viewer is in the giantess' hand. I tried to train the prompt "inHand" to be more exclusive to this position (whereas "arm" could be vague), though "arm" tends to be a better prompt for this anyway. I might use openpose to produce a bunch of training images to improve this.from above/high angle: fairly reliably places the camera above the character with her looking up, as if the viewer is dangling above her mouth.various camera distances - as you can see from the grids, trying to directly prompt the camera distance has mixed success. I've generally found that if you specify extreme close-up, the camera will go as close as it can, but it is still biased in favor of including clothes and body parts specified in the prompt. As stated above, (extreme close-up:2) can make the mouth dominate the frame and you can specify things like skin tone and face shape with some success.closed mouth - the LoRA doesn't love rendering a closed mouth, but with the facial expressions it prefers, a toothy grin with this LoRA can produce something out of a pre-hardvore nightmare. I might add some images to train on this better. Combine with a crazed expression or yandere LoRA and you've created a titan.prey - intended to render people in or near the mouth. I didn't have high hopes for this one, given SD's difficulty with rendering multiple subjects of vastly different sizes and in unusual poses. In cases of extreme close-ups, sometimes I can force it, but it tends to scramble the whole image. I'm working on a dedicated LoRA that might have more luck.

Files with SHA256: 4920fd35fb70b0a51f6ec2fb2524e4270f08a0ee774965d8afa015c4b70252d2