LoCon trained on @dragonsitoo3o's (deleted) artstyle for Illustrious modelsImages without a model tagged are done with cocoStyleNoobaiXL_v70, a model which is no longer on CivitAI<h2 id="general-usage-notes:">General usage Notes:</h2>For more specific issues check the "about this version" section of the specific model.Thus far my most successful permanent positive/negative combo is:<pre><code>masterpiece, best quality, good quality </code></pre><pre><code>lowres, worst quality, low quality, oldest, artist name, signature, patreon</code></pre>This likely changes depending on settings and merges, but I've not needed to modify them very much.Strength depends on merge lora compatibility and what you're trying to do, generally:<ul><li>Base Illustrious: Not recommended</li><li>Compatible merges: 1</li><li>Poorly Compatible merges: ~.6</li></ul>This is not my finest training and I find that I need to do a lot of cherry picking, but I like this style so ¯\_(ツ)_/¯.Issues of particular note:<ul><li>The eyes tend to be inconsistent or heavily artifacted</li><li>Fine details have more of an AI/Unnatural appearance than normal</li><li>The closer you prompt the bg to the artist's images (simple/lined/polka dot backgrounds, borders) the more likely you are to get little doodads around the subject (hearts, motion lines, notice lines) or outlines</li><li>Occasionally may add an anklet or a choker to the character.</li></ul>The sketch mannequin on top of the character is part of the style.<h2 id="general-workflow-for-sample-images-(on-comfyui):">General Workflow for Sample Images (on ComfyUI):</h2><ul><li>Sampler from images is done with nothing special</li><li>No special scheduler is used</li><li>Conditioning is split (A1111 BREAK) into 3 parts, generally BG and camera/Subject Face and attributes/Subject body and clothing</li><li>Initial resolution is selected from the SDXL whitepaper appendix I: <a target="_blank" rel="ugc" href="https://arxiv.org/pdf/2307.01952">https://arxiv.org/pdf/2307.01952</a></li><li>Image is upscaled by 1.5x using Lanczos resampling</li><li>Second sample pass is the same, but at around 15 steps at .4 denoise</li><li>A facedetailer is used after the upscale</li></ul>

Files with SHA256: 0aca0c737245700b47085184760ad6cee0be207f865997a39227a39039a49d66