<h2 id="heading-682">⚠️ This LoRA is a WIP. Don't expect good results. ⚠️</h2><p>Trigger word:</p><pre><code>scp096</code></pre><p>I don't really know what tags or models it works best on, you're free to try and experiment. I'm looking forward to see your creations!</p><p></p><h3 id="heading-9">This LoRA is based of the SCP-096 model found in the game <a rel="ugc" href="https://scpslgame.com/"><strong>SCP: Secret Laboratory</strong></a>.</h3><p></p><h3 id="heading-2">How I made this:</h3><p>I approached this one a bit different. I wasn't satisfied with images of SCP-096 I found online, so I asked a friend <em><s>(GRACIAS FISCAL CTM)</s></em> to join a <strong>SCP: Secret Laboratory</strong> server so I could take pictures of him as SCP-096. I got around 70~ images (both online and the ones I took).</p><p>I used CLIP interrogation inside Automatic1111 stable difussion web ui to tag all 70~ images because Deepboru interrogation would get maximum 3 tags (nude, standing, solo) which wasn't enough. (And I really don't know how to tag these kind of photos with danbooru tags or similar).</p><p>Also, CLIP interrogation would get some things wrong (like "a man holding a toothbrush") so I described some images myself.</p><p>I'm getting a lot of the in-game background scenery (which I DON'T like). <em>(In the showcase post, I typed "in the snow with his eyes and mouth wide open, day, clear blue sky, mountains" in the prompt, but I'm still getting some parts from the actual game, like the street or railings).</em></p><p><br />I don't think the dataset was bad, but I could really use some help on how to improve this LoRA.</p><p></p><h3 id="heading-683">I really want to update this LoRA to make it better, but I'm still learning so I could use ANY help. Thanks and have fun!</h3>

Files with SHA256: 48ed29b05076d6d5d80478f4c3b28a5e0b77fc78ec767e8d7eb1e732394dd249