<h3 id="wa-a-very-awa-locon-extract-of-a-noobai-xl-finetune.">Wa, a very awa LoCon extract of a NoobAI XL finetune.</h3>Trained on NoobXL vpred, but compatible with other Illustrious/NoobAI-XL based mixes.The finetuned model was trained using a select number of artists. The motivation behind this is to improve the aesthetics of the noob series of models, specifically on the anatomy and manga side of it.I started training the noob model because I found that the style of some of my preferred artists could not be well replicated and the finetune kinda evolved from there. As of now, the dataset includes high-quality work from a total of 9 artists, and more might be added in the future.Recommended settings<ul><li>Negative: <code>lowres, worst quality, low quality, watermark, expressionless, no pupils</code></li><li>Normal (~1MPx): Euler SMEA Dy CFG++ (1.5 cfg strength) &amp; Beta/KL Optimal</li><li>Hires (~3MPx): Kohaku_LoNyu_Yog (4.5 cfg strength, 0.4-0.7 denoise strength) &amp; KL Optimal (warning: kohaku sampler is quite slow, but is more consistent for hires)</li><li>Hires (~7MPx): Kohaku_LoNyu_Yog (4.5 cfg strength, 0.25-0.37 denoise strength) &amp; KL Optimal</li></ul>Examples (no artist tags)<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/d699f14e-983d-4dd1-a82e-7b20d570143c/width=525/d699f14e-983d-4dd1-a82e-7b20d570143c.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/4e8ed12f-003f-4dd5-82c2-3d2531bbc33e/width=525/4e8ed12f-003f-4dd5-82c2-3d2531bbc33e.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/c6c30dd9-90a3-4425-a37a-0487ad849805/width=525/c6c30dd9-90a3-4425-a37a-0487ad849805.jpeg" />Changelog:<ul><li>v12.2 (experimental) - NoobAI XL 1.0&amp;v29b vpred mix</li><li>v11.3 - NoobAI XL 0.75S vpred</li><li>v7.1 - NoobAI XL 0.65S vpred</li></ul>

Wa LoCon (NoobAI-XL)

Note: this is model was trained with experimental settings and might not work too well with all noob mixes; you can try reducing the strength of the lora or using a cfg++ sampler if you're encountering issuesTrained on NoobAI-XL 1.0&amp;v29b vpred mix<ul><li>967 manually selected and cropped images (+443 regularization images)</li><li>EVA2 tagger</li><li>24GB VRAM config (7900 XTX)</li><li>EDM2 (easyscripts only)</li><li>Tag dropout rate 7.5%</li></ul>Artists:<ul><li>damu ningen (previous version used "dam")</li><li>nonco</li><li>michiking</li><li>jp06</li><li>try (lsc)</li><li>seto yuuki</li><li>homunculus (alchemynote)</li><li>kisaragi gunma</li><li>zero (miraichizu)</li></ul>

Files with SHA256: d479279a3d41ed40dcc0da827958ef4b6fa5fb96784741f138608c3a03ca3575