This LoRA generates the goat girl/succubus which comes from this single source image: <a target="_blank" rel="ugc" href="https://doodleordie.com/c/uH576FLjb#4">https://doodleordie.com/c/uH576FLjb#4</a>Works best around strengths 0.6-0.8, but it does okay up to 1.0 and below 0.6 it makes similar characters.Trained model is Toonify but it works fine with many models.Trigger word: ws goat womanAssisting words: ws skin, ws breasts, ws horns, ws hair, ws eyes, ws goat ears, ws hoop earring, ws lipstick, ws tail, ws star face tattooYou don't need to include "ws" on each of these, though the effect may be stronger if you include it. Captioning it this way associates these captions more strongly with the trigger word and so they're softly on by default. Adding them will make them much stronger.ws skin - red skin (without mentioning the color red)ws breasts - ridiculously huge breasts, especially stratospheric with "huge breasts" or hyperfusionws horns - red horns with the specific shape (without mentioning red)ws hair - the light green aqua hair of the character (without mentioning light green aqua)ws eyes - yellow eyes with a purple or red ring (without mentioning this coloring)ws goat ears - the distinctive goat ears of the character, differs slightly from normal demon earsws hoop earring - a simple golden hoop earringws lipstick - black/blue lipstick (without mentioning these colors)ws tail - a demon tail (without mentioning demon)ws star face tattoo - a black star tattoo on the face, works best close upYou can pick and choose which elements you use.Other relevant captions: cleavage, choker, black shirt, arms bent, hands on breasts, breasts on table, smile, speech bubble, text on shirt, simple background, close up, back of headThe entire dataset uses "simple background" so if you want a background it's highly suggested that you include it as a negative prompt.This model is intended to be used with high res fix at a scale of 2x. It has 2 native resolution pairs: 520x392 -&gt; 1040x784 and 512x512 -&gt; 1024x1024. That said, it seems to work fine over a wide range of resolutions.

Woot Succubus

Initial public version, previous versions were not reasonably flexible and had much poorer captioning.The included training dataset omits the original artwork and simple edits, which was used for 00-04 and 11. I upscaled it by 2x using R-ESRGAN 4x+ Anime6B for 00, and then painted out various elements for 01-04. Then later I did a 4x upscale and made a close up crop of 1040x784 around the face for 11.

Files with SHA256: a6bae0647cc4d97d2049a8ceaedcb947b4548b354e341441e3398b5f3779f03f