<h2 id="overview">Overview</h2>This model is trained to understand the concept of wings attached to a humanoid character's back. The base model does have some understanding of this concept, however it has a tendency to misplace them if the subject is facing away, facing to the side, or otherwise not perfectly facing the camera; this LoRA aims to grant understanding of wing placement regardless of subject orientation as well as solidify the definition of different wing concepts.There are definitions for three wing types: Bat Wings (which are defined by a bony structure joined by flesh), Butterfly Wings (which are defined by a pair of flat wings with a large top segment and a smaller bottom segment, generally with a black structure around colorful portions), and Feathered Wings (which are birdlike, and naturally are covered in feathers, much like the traditional depiction of angel wings).Any of the wings can be triggered by notating in the prompt that the subject in question "has &lt;type&gt; wings," such as<pre><code>The person has bat wings
the man has feathered wings
the woman has butterfly wings</code></pre>For all wings, the default behavior will be the wings resting in a somewhat neutral or retracted position, but if you want them spread wide, mention that the wings are "fully extended;" you can also specify different variant of extension, such as "partially extended." For bat and feathered wings, more monstrous variants can be triggered by mentioning that the wings are "jagged." For example:<pre><code>the man has jagged black and white feathered wings fully extended</code></pre><h3 id="butterfly-wings-variants">Butterfly Wings Variants</h3>Butterfly wings are a more general appearance and don't have a specific species as the default wing type. There is a notable degree of monarch butterfly wing patterns in the training data, which can be prompted by mentioning that they are monarch wings, for example:<pre><code>the woman has orange monarch butterfly wings</code></pre>Butterfly wings also have training for "transparent" as a color type, which will still show the dark wing structure but have the colorful iridescence portions be transparent.<pre><code>the person has transparent butterfly wings</code></pre>There is also understanding for the butterfly wings to glow, which will cause the colorful portions to have a glow, but leave the dark wing structure normal, adding a stark contrast for the colors.<pre><code>the person has glowing yellow butterfly wings</code></pre>If you don't want the black structured detailing on the wings, you can mention that the wings are "simple."<pre><code>the person has simple green butterfly wings</code></pre><h3 id="feathered-wings-variants">Feathered Wings Variants</h3>Feathered wings' default appearance will be similar to the ones seen on the average bird (flat, flush with contours of the wing), however you can trigger a more downy version of the wings by notating that they are "fluffy."<pre><code>the person has white fluffy feathered wings</code></pre>California AB 2013 Training Data Disclosure<ul><li>This LoRA was fine-tuned using visual data consisting of image pairs. The training data may include copyrighted material owned by third parties. No training data was licensed or purchased. This LoRA is provided for non-commercial use only under the terms of its distribution.</li><li>The total dataset consists of 358 images in total, which are a mix of synthetic data and images that were posted to publicly available locations online. This is the total used for training after low quality sets were culled. Dataset was collected/created in 2026.<ul><li>Bat Wings: 78 image pairs (156 images in total)</li><li>Butterfly Wings: 36 image pairs (72 images in total)</li><li>Feathered Wings: 65 image pairs (130 images in total)</li></ul></li><li>Data was processed through standard resizing, cropping, normalization, and labeling steps. Synthetic images were included as part of the training dataset.</li><li>This model is intended for non-commercial, experimental, and educational use. Generated outputs may reflect copyrighted visual styles or themes associated with the underlying training data. Users are responsible for ensuring compliance with applicable copyright law, other intellectual property laws, and all other applicable laws.</li></ul>

Files with SHA256: 4ec7165fa8f76b125c2281f971a2e105e3c7aecf23518aad9a761c8cb8af45dc