<h3 id="raehoshi-illust-xl">Raehoshi Illust XL</h3>an enhanced iteration built upon the Illustrious XL model. It aims to elevate the visual style by addressing some of the limitations in the original, such as oversaturation and artifact noise. While these issues are not entirely eliminated, noticeable improvements have been made. The goal is to deliver a more polished, balanced output while staying true to the strengths of the base model.<hr />Why Early Access?Early access helps keep the project going. I don’t have my own GPU, so all training is done through rented cloud GPUs and that gets pretty expensive. By getting early access, you’re directly supporting the development of my models and helping me keep improving them. If you'd like to support me further, you can also <a target="_blank" rel="ugc" href="https://ko-fi.com/raelina">buy me a coffee on Ko-fi!</a> Every bit of help means a lot and keeps the future updates coming.<hr /><h3 id="recommended-setting">Recommended setting</h3>Positive prompt :<pre><code>masterpiece, best quality, very aesthetic, absurdres</code></pre>Negative prompt :<pre><code>bad quality, worst quality, jpeg artifacts, sketch, bad anatomy, signature, watermark</code></pre>Steps : 25+CFG : 5-7Sampler : euler a or dpm++2m karras (euler for vpred)Standard resolution :<pre><code>832 x 1216, 1216 x 832, 1152 x 896, 896 x 1152, 1344 x 768, 768 x 1344, 1024 x 1024</code></pre>High resolution :<pre><code>1024 x 1536, 896 x 1536, 1536 x 1024, 1536 x 896</code></pre>Hires.fix Setting:<ul><li>Upscaler : <a target="_blank" rel="ugc" href="https://openmodeldb.info/models/4x-Remacri">4x Foolhardy Remacri</a></li><li>Hires step : 10-15</li><li>Denoise : 0.1-0.3</li></ul><h3 id="special-tags">Special Tags</h3>Quality Tags:<ul><li>masterpiece</li><li>best quality</li><li>good quality</li><li>average quality</li><li>bad quality</li><li>worst quality</li></ul>Rating Tags:<ul><li>safe</li><li>sensitive</li><li>nsfw</li><li>nsfw, explicit</li></ul>Aesthetic Tags:<ul><li>very aesthetic</li><li>aesthetic</li><li>displeasing</li><li>very displeasing</li></ul><h3 id="training-details">Training Details</h3>The model was developed using a two-stage fine-tuning process. In Stage 1, new series and characters were introduced into the model. Stage 2 focused on fixing issues and enhancing the overall style for improved output.Stage 1<ul><li>Dataset : v1-31k, v2-37k, v3-34k, v4-60k, v5_v5.1-18k, v6-15k, v7-39k, v8-41k, v9-30k, v10-30k, v11-29k</li><li>Hardware : 2x A100 80gb, v3, v4, v5, v5.1-2x H100 80gb, v7,v8, v9, v10-RTX PRO 6000</li><li>Batch size : 32</li><li>Gradient accumulation steps : 2</li><li>Learning rate : 6e-6</li><li>Text encoder : 3e-6</li><li>Epoch : 15</li></ul>Stage 2<ul><li>Dataset : v1-2.5k, v2 and v3-2.3k, v4-2.5k, v5-2k, v5.1-1.8k, v6-1.5k, v7-1.7k, v7.1,v8-4.1k, v9-1.9k, v10-2.4k, v11-3k</li><li>Hardware : 1x A100 80gb, v7_v7.1,v8, v9, v10-RTX PRO 6000</li><li>Batch size : 48</li><li>Gradient accumulation steps : 1</li><li>Learning rate : 3e-6, v5.1-2.5e-6</li><li>Text encoder : disable</li><li>Epoch : 15</li></ul><h3 id="list-of-new-seriescharacters-trained:">List of New Series/Characters Trained:</h3><ul><li>Zenless Zone Zero</li><li>Wuthering Waves</li><li>Honkai: Star Rail</li><li>Genshin Impact</li><li>Arknights: Endfield</li><li>Umamusume</li><li>Azur Lane</li><li>Arknights</li><li>Fate/GO</li><li>Dandadan</li><li>Make heroine ga oo sugiru</li><li>Kusuriya no Hotorigoto</li><li>Hololive from justice and dev is</li><li>Indie Vtuber Dooby, Yuuki Sakuna, Nimi Nightmare, and S***</li><li>100 girlfriends who really love you</li><li>Haite kudasai takamine-san</li><li>Alina clover</li><li>Nikke: bready and little mermaid</li><li>Kpop Demon Hunters</li></ul>Full character list are available <a target="_blank" rel="ugc" href="https://rentry.org/xbbfsdua">here</a>For character trait details prompts, please refer to the Danbooru site for accurate tags and references.<h3 id="license">License</h3><a target="_blank" rel="ugc" href="https://freedevproject.org/faipl-1.0-sd/">Fair AI Public License 1.0-SD</a><h1></h1>Special thanks to Joe for supporting my worksSpecial thanks to Juno for supporting my works and help me with early tester

Raehoshi illust XL

<ul><li>Training Pipeline Upgrade: Transitioned to my <a rel="ugc" href="https://github.com/Raelina-Rae/diffusion-pipe">custom fork of diffusion-pipe</a> framework for enhanced stability and efficiency.</li><li>Dataset &amp; Character Overhaul: Shifted dataset curation strategy to a release-date-based timeline (up to the June 2026 knowledge cutoff) rather than focusing strictly on isolated character targets. This comprehensive update significantly improves overall fidelity, clothing accuracy, and aesthetics for all previously supported characters.</li><li>Advanced Training Optimizer: Implemented a new Warmup-Stable-Decay learning rate scheduler during training.</li><li>Expanded Knowledge Base (Up to June 2026 Cutoff): Comprehensive dataset expansions for the latest characters, outfits, and visual assets from the following titles:<ul><li>Wuthering Waves</li><li>Zenless Zone Zero</li><li>Honkai: Star Rail</li><li>Genshin Impact</li><li>Neverness to Everness</li><li>Arknights: Endfield</li></ul></li></ul>

Files with SHA256: 0753c9ddcb82f4d47d98ca2f65b0e5944a8e4528b848d0741654be0740973e64