<p>This LoRA is designed to generate a viewpoint in which the character's penis, torso, and face fit within the frame, making it appear as though the viewer is looking up while performing fellatio on that character.</p><p>The dataset consists of 108 images generated using ControlNet to control the generation process. Each image was generated using line art that I manually composited in Photoshop. The goal is to achieve consistent angle and composition across a various female/futanari body types.</p><p>While it was entirely trained on futanari-focus images, it could likely be used to generate male-focus images as well. Although, I'm not sure the composition would be as consistent (need further testing). </p><h3 id="trigger-prompts:"><span style="color:rgb(21, 170, 191)">Trigger Prompts:</span></h3><pre><code>fellatio taker pov, standing, </code></pre><pre><code>fellatio taker pov, sitting, </code></pre><h3 id="customization-tags:"><span style="color:rgb(21, 170, 191)">Customization tags:</span></h3><pre><code>arms behind back, outstretched arm, outstretched arms, saliva on penis, ejaculation, cum on penis, veiny penis</code></pre><h3 id="tips-and-known-quirks:"><span style="color:rgb(21, 170, 191)">Tips &amp; Known Quirks:</span></h3><ul><li><p>In the prompt, you can omit the following tags: <code>taker pov, implied fellatio, straight-on, from below, close-up, foreshortening</code>. As these tags were used to generate each image in the dataset, but omitted during the training process. So in theory, the model should essentially recognize the trigger words: <code>fellatio taker pov</code> as these tags.</p></li><li><p><strong><em>incoming after more test...</em></strong></p></li></ul><p></p>

Fellatio Taker POV (Sitting, Standing, Futa)

<p>Trained on Illustrious XL 2.0, 108 images, 832x1216.</p><p>Network dim16/alpha16, UNet LR 0.0002, Min SNR gamma 4, Noise offset 0.05, AdamW8bit, 2,000 steps cosine.</p>

Files with SHA256: 97fa9f7d876f7e22f962047c5d10cf50cf0f20df669d76a77770491c253fe190