<h1>⭐ StarIllustrious</h1>
<p><em>Starlight, soft glow, and a sky full of wishes.</em></p>
<p>StarIllustrious is an <strong>Illustrious</strong> (SDXL) merge for luminous anime illustration: deep night-blue palettes, glowing highlights, clean faces and expressive, sparkling eyes. It blends the rich colour and strong prompt following of <strong>WaiSnackMix Collection v1.1</strong> with the soft 2.5D depth and lighting of <strong>Luminelle v1.0</strong>, with nothing else baked in.</p>
<p>It shines on portraits and upper-body shots, starry skies, glowing magic and anything that catches the light.</p>
<p><strong>No trigger words.</strong> Prompt with Danbooru tags.</p>
<hr />
<h2>🌠 The recipe</h2>
<ol>
<li><strong>Base (A):</strong> <a rel="ugc" href="https://civarchive.com/models/2430439">WaiSnackMix Collection v1.1</a> by MGHerder</li>
<li><strong>Blended in (B):</strong> <a rel="ugc" href="https://civarchive.com/models/2961356">Luminelle v1.0</a> by VIA_AI, with SLERP at t = 0.45 (roughly 55 % A / 45 % B)</li>
</ol>
<p>SLERP blends along the arc between the two models instead of averaging in a straight line, so neither parent washes the other out. The text encoders and VAE come from the base. No LoRAs were added. Merged and converted locally with stable-diffusion.cpp.</p>
<h2>✨ Settings</h2>
<ul>
<li><strong>Steps:</strong> 30 (25–35 works)</li>
<li><strong>CFG:</strong> 5. Stay between 4 and 6; higher tends to overbake the image.</li>
<li><strong>Sampler / scheduler:</strong> Euler a + Karras</li>
<li><strong>Resolution:</strong> 832×1216 portrait, 1216×832 landscape, 1024×1024 square</li>
<li><strong>Negative prompt:</strong> <code>worst quality, low quality, lowres, bad anatomy, bad hands, extra fingers, jpeg artifacts, blurry, artist name, signature, watermark</code></li>
<li><strong>VAE:</strong> baked in</li>
</ul>
<h2>🔭 Prompting</h2>
<p>StarIllustrious reads Danbooru tags, not sentences:</p>
<ul>
<li>Comma-separated tags, most important first: <code>1girl</code> / <code>1boy</code> / <code>no humans</code>, then hair, eyes, outfit, expression and pose, then the setting, lighting and framing.</li>
<li>Put the quality tags at the end: <code>masterpiece, best quality, absurdres</code>. Pony-style score tags do nothing here.</li>
<li>Framing tags such as <code>portrait</code>, <code>close-up</code>, <code>upper body</code> and <code>cowboy shot</code> work well; it is at its best close in.</li>
<li>Tags whose names contain brackets need them escaped, like <code>star \(symbol\)</code>, or they are read as weights.</li>
<li>For mature content, add <code>child, loli, aged down</code> to the negative and <code>mature female</code> to the positive.</li>
</ul>
<p>More tips in <a rel="ugc" href="https://civarchive.com/articles/23210">Arctenox's simple prompt guide for Illustrious</a>.</p>
<h2>🪐 Which file?</h2>
<ul>
<li><strong>FP16</strong> · <code>starillustrious-v1.0-fp16.safetensors</code>: the full checkpoint (UNet, text encoders and VAE). Every showcase image was rendered with this file.</li>
<li><strong>FP8 (scaled)</strong> · <code>starillustrious-v1.0-fp8_scaled.safetensors</code>: a lighter checkpoint for ComfyUI.</li>
<li><strong>Q8_0 / Q5_K / Q4_K</strong> · <code>starillustrious-v1.0-q8_0.gguf</code> and friends: all-in-one GGUFs for stable-diffusion.cpp. ComfyUI-GGUF's UNet loader reads only the UNet from them, so take the text encoders and VAE from the FP16 file.</li>
<li><strong>INT8</strong> · <code>starillustrious-v1.0-int8.safetensors</code>: for ComfyUI INT8 (convrot) workflows. It also runs natively in recent stable-diffusion.cpp builds.</li>
<li><strong>W4A4</strong> · <code>starillustrious-v1.0-int4_convrot.safetensors</code>: ComfyUI INT4 W4A4. Experimental, and it does not run in stable-diffusion.cpp yet.</li>
</ul>
<p>Every file is converted directly from the FP16 master, never from another quant.</p>
<h2>🌌 Content note</h2>
<p>StarIllustrious handles mature content when you prompt for it with <code>nsfw</code>. The last two showcase images contain nudity; everything else is SFW. All characters are adults. Please follow CivitAI's rules and the laws where you live.</p>
<h2>☄️ Cover prompt</h2>
<blockquote>
<p>1girl, solo, long hair, silver hair, floating hair, blue eyes, star-shaped pupils, star hair ornament, white dress, looking at viewer, close up, parted lips, night sky, starry sky, milky way, shooting star, light particles, wind, upper body, from below, masterpiece, best quality, absurdres</p>
</blockquote>
<p>30 steps · CFG 5 · Euler a / Karras · 832×1216 · seed 695841494</p>
<p>Every showcase image keeps its full prompt and settings, so you can remix any of them.</p>
<h2>💫 Credits</h2>
<p>Thank you to <strong>MGHerder</strong> (WaiSnackMix Collection), <strong>VIA_AI</strong> (Luminelle), and <strong>OnomaAI</strong> for Illustrious XL. StarIllustrious would not exist without their work. 💙</p>

StarIllustrious

<p><strong>v1.0: first star</strong> ⭐</p>
<ul>
<li>SLERP merge of WaiSnackMix Collection v1.1 (A) and Luminelle v1.0 (B), t = 0.45</li>
<li>No LoRAs baked in; text encoders and VAE from the base</li>
<li>Seven formats: FP16, FP8 (scaled), Q8_0, Q5_K, Q4_K, INT8 and W4A4 (INT4 convrot), all converted from the FP16 master</li>
<li>Recommended: 30 steps · CFG 5 · Euler a / Karras · 832×1216</li>
</ul>

Files with SHA256: 81335a06484430044560196e79433b1dc3f6ca33ae11997174502a0307709c1a