This is my first attempt of making a LoRA. I know that another Beidou LoRA already exists on the site, but i wasn't satisfied with it results, hence i decided to make one myself. I want to thank @Hollowstrawberry for the collab he made. And also @rockerBOO, @ChameleonAI, @zykurv, @IndolentCat for their tips and tricks! The reviews and feedback will be very valuable for me. My goal was to make it very versatile and accessible for simple prompting (you will get better results with high prompting skills), so you can use any outfit or any hairstyle whatever your creativity and fantasy could lead to. This LoRA is capabable of making NSFW if you want to.<h3>The main trigger word:</h3>is <code>beidouep</code>. To recreate Beidou's original outfit you will need some extra triggers such as: start with <code>ogoutfit</code> that's was my main idea how to separate generations for variety; <code>hair stick</code> — her unique hairpin, i don't recommend using any other words to describe it; <code>fingerless gloves</code>, <code>fur trim</code>, <code>thigh boots</code>, <code>pelvic curtain</code>. Those are main components of her original outfit and you can use them in negative prompt as well. You can also try adding <code>cleavage cutout</code> if it's not showing up or you can add that to negative prompt if you don't want it.<h3>Next to her eyepatch:</h3>it always will be on different eyes, this is because training done with flipped images and also some of her fanart is just made with wrong eyepatch placement. If eyepatch is not showing up (this can happen on some models), I found an interesting trick during testing: you can add <code>one eye covered</code> to prompt and 98% of the time it will show up. On some models it works fine without it, but if you seeing that you current model struggling with it, add it to prompt.<h3>Weights:</h3>This LoRA was trained on AnyLora from @Lykon, so sweet spot for it is 0.7, however depending on the model you using you might want to adjust that. Especially if you will be trying to use it with realistic models. I think it works fine with semi-realistic models, I have not tried it with realistic models.I'm not sure what name LoRA will get on downloading, so please make sure you using the proper name in prompt of the file that you downloaded if you going to copy prompts from previews.<h3>Negative embeddings:</h3>Two negative embeddings were used in preview prompts (pay attention to weights): <a target="_blank" rel="ugc" href="https://civarchive.com/models/7808/easynegative">https://civarchive.com/models/7808/easynegative</a> <a target="_blank" rel="ugc" href="https://civarchive.com/models/16993/badhandv4-animeillustdiffusion">https://civarchive.com/models/16993/badhandv4-animeillustdiffusion</a>

Files with SHA256: 29895e433d667e5456482056299ccf0783fa230f025c76e85b0c508e01c1fea6