Works great with <a target="_blank" rel="ugc" href="https://civarchive.com/models/22220/cardos-animated">CarDos Animated</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/7371">ReV Animated</a>.A LoRA model of Luna, the character of the Archer class from the now-defunct mobile game Darkness Rises developed by Boolean Games and published by NEXON. The character in question was modeled by 3D artist Soo Hwang.The dataset was comprised of 64 renders covering most of the angles possible of her in-game model made by myself using Blender, and 42 extra images from both development behind-the-scenes footage and promo screenshots, all properly touched up and with no JPEG compression. My renders were mostly focused on her face and upper body with different hair models, so the range should be very good when it comes to that.Please use the "darknessrises-luna" trigger word.Her base costume is made out of a frilled bra and panties. If you keep getting that in your generations, try adding (frilled bra, frilled panties:1.4) to your negative prompt.A good chunk of the images had a grey gradient background. If you keep getting that in your generations, try adding (grey background, simple background, gradient background:1.4) to your prompt.These two scenarios shouldn't happen often if you specify what you want for clothing and background beforehand, but still, I'll leave this info here just in case.While I do recommend CarDos Animated as it was the one I used to make the previews, I do encourage you to try this out with other models. It kinda works with AOM3, but you need to make your prompt steer into a more rendered look for her features to be well-defined.Something I also noticed is that using (3d, photorealistic) makes her face stick closer to her original proportions. I didn't specify an art style during training so I have no idea why this is the case, but apart from a single image (the one where the hoodie is halfway over her head), none of the preview images had those words.If you're having trouble getting it to look right and the LoRA is overstepping its boundaries into the rest of the scene, here's what I usually do:<ul><li>Generate a base image at 512x with the LoRA weight set to 0.3</li><li>Inpaint just the face with the weight set to 0.6-0.8</li><li>Upscale with a low denoising value and the weight set to 0.3 again.</li><li>Keep inpainting with different denoising and weight values until I get the look I want.</li></ul>You can either let the face keep the "gamey" look or let the model run freely over it while retaining the overall shape of her face, but you'll need to know your way around inpainting, denoising, and all that good stuff for that.

Files with SHA256: 71c9387c5d1bd60bb3b28d5e59312b84b3d94c5aaa3dffaad4fde3cb8833a633