<p>My grand experiment rather than merge models together like with <a target="_blank" rel="ugc" href="https://civarchive.com/models/2831965/blaisn-lovely-gt">Blasian Lovely </a>I generated images with a mix of loras and used those images as the dataset for this lora. Training data was made with a mix of <a target="_blank" rel="ugc" href="https://civarchive.com/models/1535380/asian-lovely-gt">Asian Lovely</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/1534225/ebony-lovely-gt">Ebony Lovely</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/1535629/latina-lovely-gt">Latina Lovely</a>. The theory works but the execution could've been better. Trigger words were trained for each mix of loras but they do not work as intended they give widely different styles so you can play around with em if you want. @s1t1n@ - 3b0t1na - B@L - Bl@s1@n. I wanna do a refined version but I had this idea when I was making the illustrious set of lovely loras and I just now got around to it over a year later so this is probably it for a moment.</p>

Files with SHA256: 05251415184631055c32ff7a5f2d2efe45446b9de6943a40e293e34f807acbab