<h1 id="heading-542">Ultra Instinct effect LoRA</h1>Making models can be expensive. Do you like what I do? Consider supporting me on <a target="_blank" rel="ugc" href="https://www.patreon.com/Lykon275">Patreon</a> 🅿️ or feel free to <a target="_blank" rel="ugc" href="https://snipfeed.co/lykon">buy me a coffee</a> ☕I need to thank Kanye from the WD Discord for this. I've always wanted to make a UI lora, but lacked data for it. Kanye made one a while ago, which was very hard to use because the dataset was only made with Goku pics from the anime (can't be helped). However he showed me how to control it in order to get something that wasn't Goku all the time.So I got a little creative. I left my PC running to generate around 5k images of various subjects overnight, then I selected around 150 good ones (mainly females) and I integrated it with anime screencaps of Goku. After training the lora I merged it with by DBS style lora at -0.44 weight to further reduce the style overfitting. This is now my second lora from synthetic data and I'm not really a fan of the process :D I don't think the results are perfect (that's why I disabled the comments) and the lora is still difficult to control, but at least it's now easy to get girls and different characters.Should work fine at around weight 1 on most models. I tested AnyLoRA, AAM, DreamShaper, NED, AbsoluteReality and even Western Ani Diffusion. Use <code>ultrainstinct, aura</code> to increase fidelity and look at the example prompts.Also remember to use ADetailer for full body images, as well as highres fix. How to use LoRA's in auto1111:<ul><li>Update webui (use <code>git pull</code> <a target="_blank" rel="ugc" href="https://www.youtube.com/embed/mn8fMF10XN4?start=31&amp;end=60">like here</a> or redownload it)</li><li>Copy the file to <code>stable-diffusion-webui/models/lora</code></li><li>Select your LoRA like in <a target="_blank" rel="ugc" href="https://www.youtube.com/watch?v=-bMeyXOZwN0">this video</a></li><li>Make sure to change the weight (by default it's <code>:1</code> which is usually too high)</li></ul>

Ultra Instinct Effect LoRA

ultrainstinct_v3_offset.safetensors

v3 offset

&lt;h1 id="heading-542"&gt;Ultra Instinct effect LoRA&lt;/h1&gt;&lt;p&gt;&lt;strong&gt;Making models can be expensive. Do you like what I do? Consider supporting me on &lt;/strong&gt;&lt;a target="_blank" rel="ugc" href="https://www.patreon.com/Lykon275";&gt;&lt;strong&gt;Patreon&lt;/strong&gt;&lt;/a&gt;&lt;strong&gt; 🅿️ or feel free to &lt;/strong&gt;&lt;a target="_blank" rel="ugc" href="https://snipfeed.co/lykon";&gt;&lt;strong&gt;buy me a coffee&lt;/strong&gt;&lt;/a&gt;&lt;strong&gt; ☕&lt;/strong&gt;&lt;/p&gt;&lt;p&gt;&lt;/p&gt;&lt;p&gt;I need to thank &lt;strong&gt;Kanye &lt;/strong&gt;from the WD Discord for this. &lt;br /&gt;I've always wanted to make a UI lora, but lacked data for it. Kanye made one a while ago, which was very hard to use because the dataset was only made with Goku pics from the anime (can't be helped). However he showed me how to control it in order to get something that wasn't Goku all the time.&lt;/p&gt;&lt;p&gt;So I got a little creative. I left my PC running to generate around 5k images of various subjects overnight, then I selected around 150 good ones (mainly females) and I integrated it with anime screencaps of Goku. After training the lora I merged it with by DBS style lora at -0.44 weight to further reduce the style overfitting.&lt;br /&gt;This is now my second lora from synthetic data and I'm not really a fan of the process :D &lt;/p&gt;&lt;p&gt;I don't think the results are perfect (that's why I disabled the comments) and the lora is still difficult to control, but at least it's now easy to get girls and different characters.&lt;/p&gt;&lt;p&gt;Should work fine at around &lt;strong&gt;weight 1 &lt;/strong&gt;on most models. I tested AnyLoRA, AAM, DreamShaper, NED, AbsoluteReality and even Western Ani Diffusion. Use &lt;code&gt;ultrainstinct, aura&lt;/code&gt; to increase fidelity and look at the example prompts.&lt;/p&gt;&lt;p&gt;Also remember to use ADetailer for full body images, as well as highres fix.&lt;/p&gt;&lt;p&gt;&lt;br /&gt;&lt;strong&gt;How to use LoRA's in auto1111:&lt;/strong&gt;&lt;/p&gt;&lt;ul&gt;&lt;li&gt;&lt;p&gt;Update webui (use &lt;code&gt;git pull&lt;/code&gt; &lt;a target="_blank" rel="ugc" href="https://www.youtube.com/embed/mn8fMF10XN4?start=31&amp;;amp;end=60"&gt;like here&lt;/a&gt; or redownload it)&lt;/p&gt;&lt;/li&gt;&lt;li&gt;&lt;p&gt;Copy the file to &lt;code&gt;stable-diffusion-webui/models/lora&lt;/code&gt;&lt;/p&gt;&lt;/li&gt;&lt;li&gt;&lt;p&gt;Select your LoRA like in &lt;a target="_blank" rel="ugc" href="https://www.youtube.com/watch?v=-bMeyXOZwN0";&gt;this video&lt;/a&gt;&lt;/p&gt;&lt;/li&gt;&lt;li&gt;&lt;p&gt;&lt;strong&gt;Make sure to change the weight&lt;/strong&gt; (by default it's &lt;code&gt;:1&lt;/code&gt; which is usually too high)&lt;/p&gt;&lt;/li&gt;&lt;/ul&gt;&lt;p&gt;&lt;span style="color:transparent !important"&gt;https://tensor.art/u/600303455797521413&lt;;/span&gt;&lt;/p&gt;

Files with SHA256: 210c8348b5699ab1b315e59311894c2bcd683fe65cfa4cdbfc00acfdf73610ca