<h1 id="heading-173">Introduction</h1>This is the SDXL version of <a rel="ugc" href="https://civarchive.com/models/9489/charactercelestine-lucullus">Celestine Lucullus LoRA/Lycoris</a>. Do not use it with any SD1.5 model. The dataset is classified using rating tags: <code>safe, general, sensitive, nsfw, questionable, explicit</code>, which can control the results.<h1 id="heading-670">Generating Settings</h1><ul><li>Sampler: DPM++ samplers, Euler a are recommended.</li><li>Eta noise delta: 31337 or 0</li><li>Upscale Method: HiresFix+<a target="_blank" rel="ugc" href="https://huggingface.co/konohashinobi4/4xAnimesharp/tree/main">4xAnimeSharp</a></li><li>Max Resolution without Upscaling: 1024x1024</li><li>VAE: No need to use external VAEs.</li><li>Embeddings: Technically no need to load any external embeddings to enhance quality, but <a target="_blank" rel="ugc" href="https://civarchive.com/models/118418/negativexl">negativeXL</a> is recommended.</li><li>Clip Skip: 1 (SDXL model does not support clip skip).</li><li>Weights: 0.6~1 are recommended (AnimagineXL may contain prior knowledge about this character, the 'oversaturation' effect of generated images can be alleviated by appropriately reducing weights of LoRA/Lycoris).</li></ul>

Files with SHA256: 7662f2435ace1539b8492dd6af27c99ffd1cb00cf66cad1fa6bab987cc35be73