<h1 id="ponyxl-version-ml893ipvs"><a target="_blank" rel="ugc" href="https://civarchive.com/models/397525/jin-avatar-the-last-airbender-ponyxl">PonyXL Version</a></h1><p></p><h1 id="base-prompt-dmr1s7ovv">Base prompt</h1><pre><code>1girl, jin, brown hair, twin braids, green eyes, green kimono</code></pre><h2 id="second-hairstyle-frni3beqn">Second hairstyle</h2><pre><code>hair bun</code></pre><p></p><h1 id="tips-and-problems-k9lpdqzjz">Tips &amp; problems</h1><p>The goal of this model is to mimic the original art style as much as possible.</p><p>So the dataset includes only original show screencaps, and since 2/3 of it happens at night and everything is dark, the generated images created may appear faded. Unfortunately, adding <code>night</code> and <code>dark</code> to the negative prompt doesn't really help. Good at generating night scenes</p><p>We can fix it by adding more synthetic images that happens at least indoors and not at night.</p>

Jin Illustrious [Avatar: The Last Airbender]

<ul><li><p>64 dims 32 alpha</p></li><li><p>new dataset: 142 images</p></li><li><p>added <code>zuko</code> tag (unstable though)</p></li><li><p><code>hair bun</code> hairstyle was replaced with <code>ponytail</code></p></li><li><p>added some good AI generations from the previous version to the dataset</p></li><li><p>added some NSFW images in dataset</p></li></ul>

Files with SHA256: f7983d13b0a561ca20b6f4cbd6b0f9871c688efbdbfc4fea4b825da846f4dfaf