[SD 1.5][LORA]Anime Screencap Style LoRA v2-30epochs
<h2>Anime Screencap Style LoRA</h2><p><strong>Making models is expensive. Do you like what I do? Feel free to </strong><a target="_blank" rel="ugc" href="https://www.buymeacoffee.com/lykon"><strong>buy me a coffee</strong></a><strong> ☕</strong></p><p>This took a lot of time, as the dataset had to be fairly large. Plus I made some mistakes along the way and had to start over a couple of times.</p><p>For the preview pics I used <strong>weight between 0.2 and 0.4 </strong>as this net is pretty strong. As for the model I used Anything V4.5, but it probably works with most <em>nai</em> based things. Trigger keyword is <code>anime screencap</code>.</p><p>30 epoch version is usually better and stronger. Still decided to add the 35 epoch because it gave me a better result <strong>once</strong>. <br /><br /><strong>Be aware that this model is NSFW and can generate explicit hentai stuff. </strong><br /><br />Also note that I have used no embedding for the examples.<br /><br /><strong>How to use LoRA's in auto1111:</strong></p><ul><li><p>Update webui (use <code>git pull</code> <a target="_blank" rel="ugc" href="https://www.youtube.com/embed/mn8fMF10XN4?start=31&end=60">like here</a> or redownload it)</p></li><li><p>Copy the file to <code>stable-diffusion-webui/models/lora</code></p></li><li><p>Select your LoRA like in <a target="_blank" rel="ugc" href="https://www.youtube.com/watch?v=-bMeyXOZwN0">this video</a></p></li><li><p><strong>Make sure to change the weight</strong> (by default it's <code>:1</code> which is usually too high)</p></li></ul>
页:
[1]