<h2 id="training-information">Training Information</h2>Trained in <a target="_blank" rel="ugc" href="https://github.com/ThetaCursed/Anima-TrainFlow">Anima-TrainFlow</a><ul><li>This LoRA is trained using Anima Base with the total image dataset of 299+ of the Elias Ainsworth in The Ancient Magus Bride.</li></ul><h3 id="promptparameter-settings:">Prompt/Parameter Settings:</h3>Lora Strength: .7 - .9<h2 id="base">Base</h2>Activation Tag:<pre><code>1boy, elias ainsworth, red eyes,</code></pre>Reinforcement Tags:<pre><code>horns, red eyes,glowing eyes, skull mask, purple skin, </code></pre><h2 id="outfit">Outfit</h2>Elias Ainsworth Outfit:<pre><code>gloves, pants, vests, dress shirt, (amulet:1.5), jewerly, </code></pre><h2 id="styles">Styles</h2>Official Style:<pre><code>official art, official style, anime screenshot, anime coloring</code></pre>3d Style:<pre><code>3d</code></pre><h3 id="after-thoughts">After thoughts</h3><ul><li>As expected the outcome is decent given it dataset size. However, I noticed when invoking for official style, there seems to be a bias of making shot low angle for him.</li><li>Nonetheless, it defaults to its official style when generating with no art tags</li><li>When integrating a 3d art style within the loRA, it seems to have failed in capturing the details of the 3d model render within the dataset. But this is more of my laziness for not adding more variety within the 3d dataset as most of it are just cowboy shot. So most 3d generation will result looking painterly than looking a 3d render.</li><li>When generating specific details, the amulet seems to be the most finicky to generate where I'm forced to emphasize with 1.5 strength or even higher.</li></ul>

Files with SHA256: 484fe413165118c1e1e30e5a2ba1af9b3306455a4d1a7d1c6c5fd44dbb4547c6