NOT STABLE YET - WORK IN PROGRESS! I'm working on training a stable model, check back later :)<h2 id="lioraharengon-v1-(illustrious-xl-2.0-lora)">LioraHarengon V1 (Illustrious XL 2.0 LoRA)</h2>Character LoRA for Liora, a female anthro harengon (rabbit-folk), trained on Illustrious XL 2.0 (EPS) for character identity consistency.<hr /><h3 id="trigger-word">Trigger Word</h3><code>lioraharengon</code><hr /><h3 id="recommended-settings">Recommended Settings</h3>SettingValueBase checkpointIllustrious XL 2.0 (EPS)FallbackNova Furry XL ILLoRA weight0.7 - 0.9 (start at 0.8)SamplerDPM++ 2M KarrasSteps28-35CFG5.0 - 5.5Clip Skip2Resolution1024x1024 or 1024x1344 (portrait)<hr /><h3 id="prompt-structure">Prompt Structure</h3>Follow the Illustrious v2.0 tag order for best results:PLAIN TEXT<pre><code>1girl, solo, lioraharengon, [rating], [pose/action], [expression], [clothing], 
[scene/background], [camera], masterpiece, best quality, amazing quality, newest, 
detailed fur</code></pre>Keep prompts under 75 tokens — beyond that, SDXL splits into batches and coherence drops.<hr /><h3 id="sfw-example">SFW Example</h3>Positive:PLAIN TEXT<pre><code>1girl, solo, lioraharengon,
furry, anthro, furry female, rabbit, grey fur, amber eyes, harengon, tail, barefoot,
standing, smile, looking at viewer,
white dress, outdoors, forest, sunlight,
full body, masterpiece, best quality, amazing quality, newest,
detailed fur</code></pre>Negative:PLAIN TEXT<pre><code>lowres, worst quality, bad quality, bad anatomy, bad hands, extra digits,
signature, watermark, oldest, human legs, plantigrade, shoes, footwear,
hair, long hair, multiple tails, multiple views</code></pre><hr /><h3 id="nsfw-example">NSFW Example</h3>Positive:PLAIN TEXT<pre><code>1girl, solo, lioraharengon, explicit,
furry, anthro, furry female, rabbit, grey fur, amber eyes, harengon, tail, barefoot,
completely nude, nipples, small breasts, pussy,
lying, on back, spread legs, blush, looking at viewer,
bed, soft lighting,
masterpiece, best quality, amazing quality, newest,
detailed fur</code></pre>Negative:PLAIN TEXT<pre><code>lowres, worst quality, bad quality, bad anatomy, bad hands, extra digits,
signature, watermark, oldest, human legs, plantigrade, shoes, footwear,
hair, long hair, multiple tails,
(clothing:1.3), (clothed:1.3), (bra:1.4), (armor:1.4), (bodysuit:1.4),
(dress:1.2), (shirt:1.2), (fabric:1.3),
partially clothed, underwear, (panties:1.3), censored, bar censor</code></pre><hr /><h3 id="template-(insert-your-own-prompt)">Template (insert your own prompt)</h3>Positive:PLAIN TEXT<pre><code>1girl, solo, lioraharengon,
furry, anthro, furry female, rabbit, grey fur, amber eyes, harengon, tail, barefoot,
[your prompt here],
masterpiece, best quality, amazing quality, newest,
detailed fur</code></pre>Negative:PLAIN TEXT<pre><code>lowres, worst quality, bad quality, bad anatomy, bad hands, extra digits,
signature, watermark, oldest, human legs, plantigrade, shoes, footwear,
hair, long hair, multiple tails, multiple views</code></pre>For NSFW, add <code>explicit</code> after <code>lioraharengon</code> and add clothing negatives:PLAIN TEXT<pre><code>(clothing:1.3), (clothed:1.3), (bra:1.4), (armor:1.4), (bodysuit:1.4),
(fabric:1.3), partially clothed, underwear, censored</code></pre><hr /><h3 id="tips">Tips</h3><ul><li>Clothing bleed: This model has strong clothing associations from training. For nude scenes, use the NSFW negative template above and lower LoRA weight to 0.7</li><li>Digitigrade legs: The model knows rabbit leg anatomy — don't over-tag it. <code>barefoot</code> in positive and <code>plantigrade, human legs</code> in negative is usually enough</li><li>Token budget: Keep prompts concise. Tags early in the prompt have more influence than tags at the end</li><li>Don't use <code>lazypos</code>/<code>lazyneg</code> embeddings — manual quality tags and negatives give better control with this model</li></ul><hr /><h3 id="known-issues-(v1)">Known Issues (V1)</h3><ul><li>Clothing tends to bleed into nude scenes (especially bra/armor on torso and side views)</li><li>Training dataset was too large (500 images) — V2 will be curated down to ~60-80 images</li><li>Some angles (3/4, side profile) are harder to get fully nude</li></ul>

Files with SHA256: c5c442f53ba4428a8ad642b144a3986b3821e3d44d2f788a49e062cbcc5f9fc9