<h1 id="ilnai-hebaron-nirtha-from-under-the-oak-tree!">[IL/NAI] Hebaron Nirtha from Under the Oak Tree!</h1><p></p><p>🥕 ~ My favorite beefy and bearded carrot head! ~ 🥕</p><p></p><h1 id="v4">v4</h1><p>Again another re-train! I learned a lot of stuff since v3 was published and decided to redo this LoRA to test things out. </p><pre><code>h3b4r0n,orange hair,short hair,green eyes,thick eyebrows,chinstrap beard,goatee,long sideburns,mature male,</code></pre><p></p><h1 id="v3">v3</h1><p>Incorporated some things I learned to this LoRA! I refined the dataset and redid all the tagging! Removed all the custom tags aside from the trigger and added the <code>anime screencap</code> tag.</p><p></p><h2 id="triggers:">Triggers:</h2><pre><code>hebar0n,orange hair,green eyes,facial hair</code></pre><h2 id="auxiliary-tags:">Auxiliary Tags:</h2><p><code>plate armor, blue cape</code> - His armor!</p><p><code>anime screencap</code> - Using this will skew generation to the style of the manhwa. You can check the comparisons I made in the gallery! First is with, second is without.</p><p></p><h1 id="v2">v2</h1><p>Went through the training data and refined it. Accuracy seems to have increased? At the very least, there are less prominent lines in generated images. A detailer pass is still required for half and full body shots (mostly to fix the face), but quality of portraits and close-ups have definitely improved even with only HiRes fix.</p><p></p><p>I've opted to tag more character traits in this version to improve flexibility. Though one downside I noticed is that sometimes, the character looks much younger if you do not include <code>mature male</code> or <code>bara</code> in the prompt.</p><p></p><p>From testing, a minimum strength of <code>0.6</code> still replicates the character well.</p><p></p><p>Overall, I'm satisfied with this result!</p><p></p><h2 id="triggers:">Triggers:</h2><pre><code>hebar0n,orange_hebar0n_hair,green_hebar0n_eyes,bara,muscular male,male focus,mature male,</code></pre><h2 id="auxiliary-tags:">Auxiliary Tags:</h2><p><code>green eyes</code> - You can use this over <code>green_hebar0n_eyes</code> to sacrifice some eye detail for more stability.</p><p><code>hebar0n_armor, armor</code> - I tagged his armor with this but the details (like engravings) are a bit lacking. You may combine it with <code>blue cape</code> which seems to help it be more accurate.</p><p><code>cross-laced clothes</code> - He will most likely default to clothes with laces when trying to generate a shirt. You can control the behavior by adding this to your negative prompt!</p><p></p><h1 id="v1">v1</h1><p>I was contemplating whether or not I should even post this LoRA since I am aware there was an existing one for Pony. However, since I mostly use IL/NoobAI and there wasn't a version for that, I decided to train one myself. I am a bit unsatisfied with the result but it does output decent images. You just <em>have</em> to use HiRes fix and a detailer pass for the eyes.</p><p>I plan to re-train this once I farm enough buzz again LOL</p><h2></h2><h2 id="notes:">Notes:</h2><ul><li><p>This is a LoRA made for my own enjoyment so quality may be variable!</p></li><li><p>The overall style gets skewed noticeably when used at max strength. Play around with strength values to offset this. I usually use 0.8 or 0.9 and just strengthen character features by adding more character trait tags like <code>beard, muscular</code>.</p></li><li><p>The backgrounds tend to become more <code>medieval</code> themed if you do not specifically describe the environment in your prompt.</p></li><li><p><code>mature male</code> tends to add more facial hair than what the character is supposed to have especially if you do a detailer pass for the face. Please add <code>mustache</code> to your negative prompt to circumvent this!</p></li><li><p>Only intended for Illustrious/Noob AI models!</p></li><li><p>Please use Adetailer to fix artifacts in eyes, hands, etc. 🙇</p></li></ul><p></p>

Hebaron Nirtha (Under the Oak Tree)

<ul><li><p>Redid dataset and tagging</p></li><li><p>Added <code>anime screencap</code> to control style bleed over</p></li></ul>

Files with SHA256: cc850cecd07cd690b52c1328dff4528175d46b0f5025afb9c2c91a2ccbaa49e6